Process Of Face Detection Cultural Studies Essay

Published: Last Edited:

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

When talking to a person, you usually look at his face; a facial expression of a person plays a very important role in communicating with others. Because of its uniqueness, the face is the most important and effective to recognize a person.

In comparison with fingerprints or retina, taking a picture of a person's face is very easy. Therefore face recognition has become one of the most popular applications in the field of computer vision (Hong et al, 1998). With the recent major terrorist attacks in the civilian world, there were more substantial interest in the development of smart cameras that can automatically detect and recognize known criminals and suspicious characters. Due to these uncertain times, humans begin to seek the support of computer systems to facilitate the process of identifying and locating faces in scenes of everyday life (Leung, 2001).

Images of faces greatly depending on the lighting, occlusion, pose, facial expression and identity. Color transformations must be implemented to deal with all the variations of color remaining in the distinctive facial skin.

Problem Statement and its Significance

The majority of the methods already developed usually have at least one of these problems: -

- Too high calculation (time, space) complexity and / or:

- Efficiency too low.

At this stage, it is necessary to emphasize that the automatic face detection, and most other objects automatic detection methods is a very pretentious task, mainly due to variations of samples that can not be easily described analytically with the parameter.

Automatic detection of human skin in two-dimensional image is a natural and complex task interesting and challenging area in computer vision. Designed to detect and locate human face region. This task is the first step in many problems such as the detection of human face and hand. Human face detection is in turn very important step in applications such as automatic identification, sex / race detection and understanding facial expressions and hand gestures. All these applications are based on the assumption that the areas of the human skin are already detected and localized.

Detection of human skin may show greater importance for the future of human / robot interaction, where a robot must first detect and locate human beings in natural images he sees. Other measures may be areas of analysis within the hand / face to detect the location of the eyes or fingers. This research studied the different methods for detecting human skin, which are based on matching model of human skin color (peer Soline, 2003).

1.3 Aims and Objectives

Objective of the research is to evaluate the process of face detection using skin color tone, so that the advantages and applications of skin color can contribute in a single algorithm. The objectives of the research are:

1. The first is to understand the different methods of face detection methods available whereby different techniques are used.

2. After analyzing the different method of an appropriate technique is proposed, which is more accurate and has a high ratio of detection.

3. Compare the proposed technique to face detection with other techniques.

1.4 Expected outcome

1. New algorithm for face detection again.

2. Develop an approach to the method of analysis of the skin area within the image based on the RGB color.

3. Publications.

The expected result of the method of face detection based on skin color is the same image that is given as input to the system, but the output image will contain just the face of the person who was the image of input.

1.5 What is Face Detection?

Face detection is defined as finding the number of an image present in face and locate their position and size. In general, the process of face detection is the sheet in two stages in the first rod of the whole image is scanned to find the area (characteristics) that can be identified as a skin color of the face, is the most common to all. And the other is the location which provides more accurate estimate of the size and position of the faces. Face detection is the first step in the identification and human recognition.

1.5.1 Challenges: Face detection method faces are different challenges such as installation, the imaging condition.

Installation: The image of a face depends on the camera relative-pose face and some facial features become partially or completely blocked.

Image orientation: Face images directly vary for different rotation around the optical axis of the camera. Orientation of the image directly affects the angle of the face.

Illumination: This problem is mainly due to the lighting, makes a big difference with the same face as compared to the difference in the different faces while comparing (Javidi.Bahram, 2002).

Occlusion: Sometimes the faces in the picture are clogged with other objects such as mustaches, beards, optical lenses and other types of objects, it is very difficult to find the exact image.

The facial expression: Some people may have a different expression at different times, which also helps to restore face detection.

Face size: Size of the face as hard to automate a face detection and recognition (Mahmood, 2006).

Condition of the image: This problem includes factors such as intensity, resolution, camera lighting, background, characteristics of image capture device and the distance between the camera and person plays a role important in the process of face detection.

1.6 Research structure

Chapter 1: Introduction to face detection covers the aims and objectives of face detection and also the problem faced in achieving effective face detection.

Chapter 2: Briefly describe the methods of face detection and explain the different techniques for face detections.

Chapter 3: Introduction to Image Processing.

Chapter 4: The proposed algorithm, Experimental Results and Comparison.

Chapter 5: Discussions, Conclusion and Future work.


Appendix - MATLAB Code



2.1 Background

Automatic detection of human skin in two-dimensional image is a natural and complex task interesting and challenging area in computer vision. Designed to detect and locate human face region. This task is the first step in many problems such as the detection of human face and hand. Human face detection is in turn very important step in applications such as automatic identification, sex / race detection and understanding facial expressions and hand gestures. All these applications are based on the assumption that the areas of the human skin are already detected and localized.

Detection of human skin may show greater importance for the future of human / robot interaction, where a robot must first detect and locate human beings in natural images he sees. Other measures may be areas of analysis within the hand / face to detect the location of the eyes or fingers. This chapter has studied the different methods for detecting human skin, which are based on matching model of human skin color (peer Soline, 2003).

2.2 Face Detection in Image

Many techniques for face detection in image were classified into four categories.

Knowledge based method

It depends on the rules for using the features of the human face. It is easy to find simple rules to describe the features of a face and their relationships. For example, a face often appears in an image with two eyes that are symmetric with respect to the other, a nose and a mouth, and has relative distance and position to represent relationships between entities. After feature detection, verification is done to reduce false detections. This approach is good for the front image, the difficulty of it is how to translate human knowledge in the known rules and to detect faces in different poses.

Face DetectionImage Based method:

In this approach, there is a predefined standard face model is used to match with the segments in the image to determine whether or not they are faces. It uses training algorithms for classifying regions face classes or not the face. The image-based techniques depends on multi-resolution window to detect faces, so that these techniques have a high detection rate, but more slowly than the basics function Eigen-faces and neural networks are examples of Techniques based on the file. This approach has the advantage of being simple to implement, but it cannot deal effectively with variations in scale, pose and shape.

Features Based method:-

This approach is based on the extraction of facial features that are not defective by variations in lighting conditions, poses, and other factors. These methods are classified according to the extracted features (Hong et al, 1998). Feature-based techniques depend on feature derivation and analysis to acquire the necessary knowledge of the faces. Characteristics can be skin color, face shape or facial features such as eyes, nose, etc. Feature-based methods are preferred for real-time systems, where multi-window scanning resolution used by image-based methods is not applicable. Color of human skin is an effective element used to detect faces, although different people have different skin colors, several studies have shown that the fundamental difference in terms of their intensity rather than their chrominance. Textures of human faces have a special texture that can be used to separate different objects. Method depends on facial features detection of facial features. Some users use the edge for detecting the characteristics of the face and then the edges together. Some others use stains and scratches instead of edges. For example, the face model consists of two blobs and three light black blobs to represent eyes, cheekbones and nose. The model uses streaks to represent the outlines of the face such as the eyebrows and the lips. Several methods using multiple features combined facial features to locate or detect faces. First find the face using features such as color, size and shape, and then verify these candidates by using the detailed characteristics such as eyebrows, nose and hair.

Template Matching method

Matched model methods use the correlation between the pattern in the input image and stored standard patterns of a set of face / face characteristics to determine the presence of features of the face or the face. Predefined templates and deformable templates may be used.

2.3 Face Detection Approaches

There are many distinct approaches to face detection (Hsuan, Krieg man, David J., & Ahuja, 2002):

1. The Top-Down Method-Based Approach. Model assumes a different face different course with fine scales. For greater efficiency, the image is searched for the coarse scale first. Once a match is found, the image is searched for the next finer scale until the finest scale is reached. In general, a single model is assumed in each scale (usually in parallel frontal view) and it is difficult to extend this approach to multiple views.

2. The Bottom-Up Feature-Based Approach: Research on the image to a set of facial features and groups in their face candidate based on their geometric relationship. Although this approach can be easily extended to multiple views, he is unable to work in conditions very different images because the image structure of the facial features vary too strong to be detected by the detectors of features.

3. In Texture-Based Approach: Faces are detected by examining the spatial distribution of the gray level information in the secondary image (using spatial gray level dependence (SGLD) matrices). It is not easily extensible to multiple angles.

4. The Neural Network Approach: Detects faces by regions different subsampling of the image with an image of a standard size, and then passing under the filter through a neural network. In general, the algorithm works very well for frontal-parallel faces, but the performance degrades when extended to different views of the face. It is not yet possible to extend the algorithm to detect faces in profile views.

5. The Color-Based Approach: Labels for each pixel according to her similarity to skin color, and subsequently qualifies each sub region as a face if it contains a large number of pixels blob of color. It can cope with different views of faces, but it is sensitive to skin color and facial shape. It should be borne in mind that the proposed method lies in this approach.

6. Motion-Based Approaches: Using the image subtraction for extracting foreground movement of the static background. The surface is then located by examining the shape or the color of the difference image. This approach does not work well when there are many moving objects in the image.

7. At Depth-Based Approach: Primary facial features are located on the basis of depth information for the face. In the first step, the stereo image pairs containing front views are sampled from the input video sequence. Then point correspondences over a large disparity range are determined using a multi resolution hierarchical matching algorithm. Finally, the facial features are located on the basis of depth information.

2.4 Color Specifications

Since the early 1990s, the cost of high-quality color cameras has become competitive with the black and white. Of course, for a human operator, color images are much more useful for identification, as clues such as hair color and complexion can be used. Assuming that the image resolution and noise performance are sufficient, the use of color images to recognize faces should provide many benefits comparable to manual identification. The color is understood as consisting of two components: luminance and chrominance.

The chrominance is the property of the object itself, which identifies the color of that object. On the other hand, the luminance is the property of the lighting environment around the object. Each color image contains both values ​​of luminance and chrominance. It is obvious that colored objects appear black on black and white in an extremely bright light. On the other hand, the glass is always transparent in all lighting conditions. It is recognized that the luminance is a major problem in the detection of human skin color from human skin color varies considerably by changing lighting conditions.

Luminance and chrominance to the colors of human skin cover a large area of the color spectrum, which makes it difficult to detect the color of the human skin on the basis of the RGB color values ​​themselves. Hence the need for the various color spaces that can reduce the variance of the skin color of man arises (Al-aqrabawi, Mohammad, & Fangfang, 2000).

2.5 Color Spaces

Color space is a method for representing color information of an image obtained. Color of human skin tends to cluster in different color spaces. Different color spaces tend to improve certain characteristics of the images at the expense of others.

The color images are usually entered as red, green and blue (RGB) and stored in computer files as two-dimensional arrays of values ​​of three different colors, color values ​​usually can not be detected by the human observer. In addition, variations in lighting highly modify the RGB values ​​of the image. Therefore, the color space is not very suitable for the detection of human skin. For this reason, we must think of another color space suitable to represent the color of human skin. It is also possible to represent the color using another coordinate system color such as HSV or YUV. HSV system is a system of polar coordinates with H denoting the values ​​of hue, saturation S and V values ​​of intensity is similar to HSV YUV with Y representing the intensity, while the UV components, also known as chrominance components, specify the hue and saturation in Cartesian coordinates. YCbCr color space can represent color images using its components Y and CbCr (Sangwine & Horne, 1998).

2.6 Human Skin Color

The color of human skin is distinctive from the color of many other objects and therefore the statistical measurements of these attributes are of great importance for face detection. Color is a prominent feature of human face. Using skin color as a primitive feature for detecting face region has several advantages. In particular processing color is much faster that processing other facial features. Furthermore, color information is invariant to face orientation. However, even under a fixed ambient lighting, people have different skin color appearance, in order to effectively exploit skin color for face detection, one need finding a feature, in which human skin colors cluster tightly together and reside remotely in background colors (Chai, Phung, & Bouzerdoum, 2001).

Many approaches in the literature use different detection procedures, either base on the RGB. Chromatic (CbCr) or Hue and Saturation (HSV) space (Al-aqrabawi, Mohammad, & Fangfang, 2000). In proposed system, the YCbCr color space is adopted since it is perceptually uniform and separates luminance and chrominance. Many research studies found that the chrominance component of the skin tone color is independent.

In YCbCr fields the brightness (luminosity) is stored as a simple element (Y) and the value of chrominance as two different elements (Cb and Cr). The values Cb and Cr represent the difference between light blue and the current calculated value, as well as the difference between red and the current calculated value, respectively. Figure 2.1 represents the chromatic distribution of the human skin area with respect to C6 and Cr values:

Cb Cr

Figure 2.1: Human skin chromatic distribution

2.7 The Skin Color as a Feature

Faces often have a characteristic color which is possible to separate from the rest of the image. Numerous methods exist to model the skin color, essentially using Gaussian mixtures or simply using look-up tables. In some studies, skin color pixels are filtered, from the sub-image corresponding to the extracted face, using a look-up table of skin color pixels. The skin color table was obtained by collecting, over a large number of color images, RGB (Red-Green-Blue) pixel values in sub-windows previously selected as containing only skin pixels. For better results, the face bounding box should thus avoid as much hair as possible (Marcel & Samy, 2001).

As is often done in skin color analysis studies (Peer & Soline, 2003), one computes the histogram H (g) of R, G and B pixel components for different face images, which is computed as the number of pixel at gray level. Such histograms are characteristic of a specific person, but also discriminate among different persons.

2.8 Skin Tone Detection & Localization of Facial Region

This work, directly locates face region boundaries based on their feature maps derived from both the luminance and chrominance of the image.

Modeling skin color requires choosing an appropriate color space and identifying a cluster associated with skin color in this space. It has been observed that normalized red-green (rg) space (Al-aqrabawi, Mohammad, & Fangfang, 2000) is not the best choice for face detection. Based on Terrill on et all's (Hsuy, Abdel-Mottaleb, Mohamed, & Jainy, 2002) comparison of nine different color spaces for face detection tint-saturation-luma (TSL) space provides the best results for two kinds of Gaussian density models (unimodal and mixture of Gaussians). Many research studies assume that the chrominance components of the skin-ton color are independent of the luminance dependency of skin tone color in different spaces.

The YCbCr color space (blue dots represent the reproducible color on a monitor) and the skin tone model (red dots represent skin color samples): (a) the YCbCr space; (b) a 2D projection in the Cb Cr subspace; (c) a 2D projection in the (Cb/Y)-(Cr/Y) subspace.

Detection skin tone based on the cluster of training samples in the Cb Cr subspace, shown in figure 2.4 (b), results in many false positives. Face detection based on the cluster in the (Cb/Y)-(C/Y) subspace, shown figure 2.4 (c), results in many false negatives. Therefore the YCbCr color space is nonlinearly transformed to make to skin cluster luma-independent. This is done by fitting piecewise linear boundaries to the skin cluster. The transformed space is shown in (figures 2.4) in which the elliptical skin model is overlaid on the skin cluster. Red dots indicate the skin cluster. Three blue dashed curves in (a) and (b), one for cluster center and two for boundaries, indicate the fitted models.

Figure 2.2: New nonlinear transformation of YCbCr color space

Figure 2.2 also shows nonlinear transformation of the YCbCr color space: the skin tone cluster shown in (a) the YCb subspaces; (b) the YCr subspace (c) the new transformed YCbCr color space; (d) a 2D projection of (c) in the transformed CbCr subspace.

2.9 Morphology

Morphology is known to many as a branch of biology that deals with the form and structure of animals and plants. Mathematical morphology is the same in the study areas and shapes in images. Morphological techniques are well developed for binary images, but many methods can be successfully extended to grayscale. For a binary image, the white pixels are usually taken to represent the regions of the foreground, while the black pixels represent background. Virtually all mathematical morphology operators can be defined in terms of combinations of erosion and dilation with set operators such as intersection and union. Operators are particularly useful for the analysis of binary images and common usages include edge detection, noise removal, image enhancement and image segmentation. The basic effect of the expansion of a binary image is to gradually enlarge the boundaries of regions of foreground (usually white pixels). Thus, areas of foreground pixels grow in size, while the holes in these regions become smaller. Erosion on the other hand, removes the limits of regions of foreground pixels. Thus, areas of foreground pixels shrink in size, and holes in the zones become larger. The technique of ultimate erosion is a particular morphological technique used to locate faces in our current system. Rather than trying to match a face model and to determine the most "shaped face" region, the ultimate erosion makes full use of the binary mapping derived from the color segmentation step to locate the center of the face. You can view the process of erosion of the islands being eroded as the rise in sea level, where the islands are represented by those of the binary image. As erosion is applied, the sea level rise a little for all neighboring pixels around each island will disappear. Logically, this means that erosion will begin by first removing the smaller islands, while the larger islands are narrowed. As the following operations are applied against erosion, the biggest island is the last surviving piece of land, and the nearest point or the center of the island will be the last to be eroded (Al-aqrabawi, Mohammad, & Fangfang, 2000).



3.1 What is Image?

A two-dimensional function f (x, y) with x and y are the spatial coordinates is called the image and the amplitude of the function f (x, y) at any point (x, y) is called the image intensity or gray level. If the value of x, y and the magnitude of f (x, y) are discrete image is said digital image.

3.2 What is Pixel?

Pixel is the basis, the smallest discrete component of the image which can be controlled. Each pixel has special place and value. They are also called pixels, picture element, and picture elements.

3.3 Image Processing:

The digital image processing term refers to the transformation of data in two dimensions (image) and digital image term refers to an array of real or complex numbers represented by a finite number of bits. Digital image processing has wide range of applications such as remote sensing satellites and spacecraft, medical treatment, treatment of radar and acoustic robotics.

There are various applications of image processing that comes from different areas of it.

The various fields of image processing are as follows.

Enhancement picture

Image restoration

Color image processing

Image segmentation

Edge detection

Histogram processing

3.4 Image Enhancement:

According to David Lindsay image enhancement is as follows: All that makes the difference if you see the darkness with light or brightness through the shadows.

The main objective is to improve the quality of images for human visualization and analysis that makes it (images) most suitable for a specific application. Increasing the contrast, blur and noise removing some of the development operation.

Image enhancement approaches are classified into two categories: spatial domain methods and method in the frequency domain. Techniques of spatial domains are based on the location of the pixels in Image and frequency domain techniques are based on the modification of the Fourier transform of an image. There are different techniques of image enhancement some of them are against stretching, power-law transformation grayscale slicing, and bit-plane slicing etc. (Rafael C. Gonzalez and Richard E. Woods, 2001)

Figure 3.4.1 Eyes (Zia-ur Rahman, 2001)

Figure 3.4.2 X-ray (Zia-ur Rahman, 2001)

3.5 Image Restoration:

Immanuel Kant:'' The things we see are not by themselves what we see that it remains completely unknown to us that the object can be themselves and outside the receptivity of our senses. We do not know, but the way we perceive things.''

The difference between the improvement of the image and image restoration is that the image improvement is a subjective process, while the image restoration process is largely objective.

Image restoration is used to fix the color, contrast, tone and repair the damage that is due to non-linear filtering produced by the sensors, the geometric distortion correction, noise filtering and limitation of the environment. Thus, restoration techniques are pointing the degradation modeling and implementation process in the reverse order to recover the original Image. Following figures will show the best result of image restoration.

Figure 3.5.1 Image restoration (ozlouisville)

3.6 Color Image Processing:

Pablo Picasso:'' for a long time, I am limited to a single color, as a form of discipline.

The use of color in the image processing is encouraged by two factors as the color often assist in the identification of objects and extraction from images is a factor and the other factor is the human eye can distinguish thousands of color shades and intensities and only two dozen shades of gray. Treatment of the color image is divided into two main parts: full color image processing and pseudo color image processing. In the treatment of color important thing to note is where the images are acquired (from the TV color camera or scanner color). And in the treatment of pseudo color, the problem is to assign color for a particular range of intensity. For many years processing color image is not used but a decade, color sensors and processing equipment of color available at reasonable prices resulting from the use of the wide range of applications such as editing, viewing and Internet. (Rafael C. Gonzalez, Richard E. Woods, 2001)

In color image processing, the color spectrum is divided into six broad regions: purple, blue, green, yellow, orange and red are formed by passing white light through the optical prism as shown below. The color in that man perceives an object is determined by the nature of the light reflected by the object. The object reflects light which is on the point of all visible wavelengths white look to the user object that promotes reflectance in a confined space of the visible spectrum has some shades of color. For object reflecting green light, for example of the order of 500 nm to 570 nm (wavelength) while observed over the wavelength of energy to another.

Figure 3.6.1 Refraction of light (Ian Flitcroft, 2007)

Characteristics of light played an important role in the science of color. The lights out of the black and white TV is achromatic light is white, gray and white are achromatic color and the light that comes out of the achromatic color is light (as below) is between 400 and 700 nm. Brightness, luminance and brightness are used to define the quality of the light achromatic.

Figure 3.6.2 Gray-scale (Tomás Castelazo, 2006)

3.6.1 Radiance: The net amount of energy that comes from the light source.

3.6.2 Luminance: It is the amount of energy from the observer perceives a light source.

3.6.3 Brightness: brightness is a subjective meaning and practice; it is not possible to measuring the brightness.

Red, green and blue are the three primary colors are separated from each other on the basis of the detection capability of cones found in the human eye. Due to the absorption property of the human eye, the colors are considered variable combination of red, green and blue which are known as primary color. All colors resulting from the combination of primary colors are called secondary colors such as magenta, yellow, cyan and red. The feature which is used to separate the colors of the other, the brightness, hue and saturation.

3.6.4 Hue:

Hue represents the dominant color as perceived by the observer, is the property associated with the color which has a dominant wavelength in a mixture of light.

3.6.5 Saturation:

It provides information on the color purity, or the amount of white mixed with the hue. For example, pink is a combination of red and white colors are less saturated. The degree of saturation and the amount of added white color are inversely proportional to another.

The combination of hue and saturation are termed chromaticity and therefore all colors are characterized by their lightness and Chroma.

Figure 3.6.3 RGB image components (Gonzalez and Woods, 2001)

3.7 Image segmentation:

Segmentation is the process of dividing an image into regions or components of the articles so that the smallest details of the image are read or analyzed. The objectives of the segmentation are change or modify the representation of the image significantly and simplified (easy to analyze). This process comes in picture when the object of interest is detected or identified from a set of objects and stops immediately when the object is isolated from the rest.

Moreover, it is unnecessary to carry out the segmentation process beyond the level of detail required identifying the object of interest. As in other areas of image processing, image segmentation has also received numerous asked some of them are: Diagnosis, study the anatomical structure of the object location in satellite images of military imaging, etc. In general, image segmentation algorithms are based on two properties of intensity values: discontinuity and similarity. In the category of discontinuity, the images are segmented according to abrupt changes in intensity. And in the second category partition images are based on the similarity exists with the predefined criteria. Based on two types of image segmentation, there are different types of approaches are analyzed among them are the most sought after edge detection and thresholding.

3.8 Edge Detection:

His approach is the detection of discontinuities in gray level. The connected set of pixels which is located at the boundary between two regions are designated edge or is the ability to measure the pixel grayscale transitions significantly and the edge detecting comprises detecting the pixels. Images are blurred sampling and imperfections bad image acquisition. Thus, the edges are more closely modeled as profile that ramp like below. The slope of the ramp and the blur in the image are inversely to another. All pixels that are located on the line are said to be ramp edge point. The length of the line which is based on the scope of the line is determined by the amount of blur. Thus, it is found that the sharp edges tend to be thin and blurry edges tend to be thicker.

In the figure below first derivative and second ramp is calculated. The first derivative is positive at the time of the transition in and out of the ramp (from left to right) and constant over the entire line of the ramp. Second derivative is positive only at the instant of transition associated with the black face, negative only at a time of transition associated with the light side and zero on the line of rail. And first derivative is used to identify the presence of an edge to a point in an image, and the second derivative is used to conclude if an edge pixel is found on the side of a light or dark image. Second derivative has two attributes, for each edge in an image; it produces two values, the other zero-crossing property. The zero crossing, an imaginary line passing through zero at the midpoint of the two values ​​of second derivative, connecting the positive and negative values.

Figure 3.8.1 Edge detection (Shehrzad Qureshi, 2009)

3.9 Histogram processing:

This is a common technique for image enhancement. This is a vertical bar graph that provides information on the distribution of the data set. When a table with a large number of measurements is given, the histogram is used for organizing and displaying data in simple format that is easy to use and user friendly. A histogram, it is easy to determine where the majority of Values ​​is within a measuring range, and the degree of variation between them.

The following example shows a part of the situation in which the histogram is necessary:

Summarize large sets of data in graphic form

Compare the results with the process specification limits

Information communicates graphically.

Help support the decision.


As shown below, the histogram consists of five parts:

1. Title: The title briefly describes the information contained in the histogram.

2. Horizontal or X-axis shows the horizontal or X axis you the extent of values ​​where appropriate measures.

3. Bars: The bars have two important features in height and width. The height represents the number of times the values ​​within the range are produced. Width represents the length of the interval covered by the bar.

4. Vertical or Y axis: Y axis is the scale that shows the number times the values ​​in an interval occurred which is termed as frequency.

5. Legend: The legend provides additional information, for example, where the data come from and how the measurements were collected.

Figure 3.9.1 Histogram Parts

3.10 Thresholding:

Thresholding enjoys a central location in applications of image segmentation because of his property and intuitive ease of implementation. It attempts to identify an object from its background. The intensity is the property that pixels in a region can share with each other. Thus, a normal way to segment these regions is a separation of light and dark regions which also known as thresholding. Product with a binary image from the gray levels by folding all the pixels below a threshold to zero and all pixels on a threshold (K.V Kale, Mehrotra, & Manza, 1999).

If g(x,y) is a thresholding version of image f(x, y) at some global threshold T.

Figure 3.10.1 Thresholding (Lindsay coome, 2005)

The problem with thresholding is that it does not take into account the relationship between pixels, but does not consider the property of the image intensity. There is therefore no Guarantee pixels identified by the thresholding process is continuous. No information is given if additional pixels are included or neglected in the region. There are different thresholding techniques among them is the simplest global thresholding where the image histogram partition using a single global threshold T. Segmentation process is then began scanning the image pixel by pixel and label each object as a background or on the basis of the gray level of the pixel is less or greater than the threshold value T. In total only a single threshold is set and the result of a problem of illumination changes through the stage causes parts to be successful in the light and in some areas darker in shade. It is one more thresholding technique based on several variables which is called as multispectral thresholding.



4.1 Introduction

In this chapter a face detection system is presented and implemented using a skin region analysis method, within the image according to RGB color, each pixel is priestly classified as being either skin or non-skin. This system is the defect the face from a background. This system is implemented using, the scientific language which is efficient to deal with matrices mathematic operation and image processing.

4.2 Color Models:

The aim of color model is to facilitate the specification of color in a certain standard. In general color model is a specification of a coordinate system, this sub-space color the space is used to represent each color in a single point.

4.2.1 RGB Color Space:

RGB is based on the Cartesian coordinate system which is an aid cube is shown below. The cube has the RGB values ​​in the three corners, colors like cyan, magenta and yellow and the other three corners, the black and white at the origin is at the corner furthest from the origin. The gray scale is located on the line joining black and white. They are called additive ''primary'' because the colors are summed to produce the desired color. Due to the high correlation between color components: red, green and blue as each component is subject to the effect of luminance of the light intensity of the environment, so that suffers dissatisfaction on the part of many image processing applications. As practical, this model is not well suited to describe colors in terms of human interpretation. (Sanjay Kr Singh, DS Chauhan, Mayank Vatsa, Richa Singh, 2003).

Figure RGB cube (Paul Bourke, 1995)

4.2.2 HSI (Hue, Saturation, Intensity) Color Space:

This color model is ideal for hardware implementation. Human practice interpretation is described in terms of Hue Saturation and Intensity (Brightness). HSI model separates the intensity component of the color information-bearing, is an ideal tool development processing algorithm based on the image description of the colors that are natural and intuitive to human health. HSI is based on the cylindrical coordinate, hue (H) is shown with a 0 angle, varying from 0 to 360, the saturation (S) corresponds to a radius ranging from 0 to 1 and Finally the intensity (I) varies along the z-axis with 0 being black and white is 1. When the value of S is 0, the color is a gray value of intensity. When the value of S is 1,

Color is the limit of the cone tope. More saturation, the color is white / gray / black. Chancing the value of the color will change the color of red that is to 00 to green which is 1200, 2400 and blue to red black in 3600. And when I = 0, the color is black and therefore H is undefined. By varying the value of I, a color can be is darker or lighter. Shaded color comes out when the value of I is adjusted while S = 1 hold.

4.2.3 YCrCb Color Space:

This color space was defined to meet the growing demand of digital processing algorithms of video information and has become widely used color space in digital videos. It has three components, two of them are the chrominance and luminance is one. This model comes in the family of television transmission color space along with YUC and YIQ are designed for space analogue PAL and NTSC (Mahmood, 2006).

YCrCb color model has been developed to allow the transmission of color information on televisions keeping in mind that the existing television in black and white still to display images in shades of gray. And has the characteristic of luminance and isolate color information, and used in many applications such as compression.

4.3 Algorithms:

4.3.1 Skin Color Based Face detection in RGB Color Space:

Skin color is the most important feature of the face and is unique because of its color ingredients. Skin color pixels can be easily detected using standard color histogram which is normalized for every future change in the intensity of luminance of the division. And thus RGB vector is transformed into a vector [r, g] color standard which in turn provides a rapid means of detecting the skin. This gives the skin color region which localizes face. As indicated early as RGB suffers from the effect of luminance but still able to allow us to eliminate some colors which are clearly outside the scope of normal skin color. After review and analysis different levels of the RGB color space, it was found that the following rule works well in eliminating some redundant pixels that are labeled as non-face (Mahmood, 2006).

0.836G - 14 <B <0.836G + 44 => Skin

0.79 g - 67 <B <0.78 g + 42 => Skin

4.3.2 Skin Color Based Face Detection in HSI Color Space:

The HSV color space is more intuitive and provides color information in a manner same way that human think of colors and how artists mix colors in general. dyed with saturation provides useful information regarding selective skin. Range for non-Hue face region is considered as shown below and it is also assumed to be the skin. HSI color model is represented using three

19 < H < 240 => Not Skin

HSI color space is defined by using the RGB as scale as transformation from RGB space to and HSI space (Gonzales and woods)

The colors are given a set of RGB color cube. The color cube is oriented so that the value component is oriented along the axis of gray, ranging from black to white. HSV are defined solely by reference to the RGB color space, they are not absolute color spaces: to specify a color precisely requires reporting not only the HSV values, but also the characteristics of the RGB space they are based, including gamma correction in use it.

Computer offers some poor cousins ​​of these perceptual spaces which can also be found in the software interface, such as HSV. They are easy mathematical transformations RGB, and they seem to be perception systems because they use the color lightness / saturation value terminology. But take a look around, do not be fooled. Perceptual dimensions of color are poorly scaled by the color specifications that are provided in these and other systems. For example, saturation and brightness are combined, so that saturation scale may also contain a wide range of light (for example, it can evolve from white to green, which is a combination of both brightness and saturation). Similarly, the color and lightness are confounded if, for example, a saturated blue and yellow can be saturated designated as the same "lightness", but have large differences in the perceived lightness. These defects make the systems difficult to use to control the appearance of a color in a systematic approach manner. If more tweaking is needed to get the desired effect, the system offers little benefit more struggling with raw RGB specifications (Cynthia A. Brewer, 1999).

Figure HSV-cone

4.4 Proposed Face Detection

The face detection system is presented to detect a face image from a background. This system is important in many applications such as face recognition to increase the speed and accuracy of recognition process that is because it deletes any undesired information. The proposed face detection system is depending on classifying the face as being either skin or non-skin.

Algorithm about Proposed Face Detection

1- Resize the original face image into (256*256)

2- Separate the RGB pallets into R,G, and B.

3- Transform the RGB space into (YCbCr)

4- Transform the (YCbCr) space into new space (YĆbĆr)

5- Skin color two detection


Ćr > 1 the pixel indicate a skin color tone


The pixel is not a skin color tone

Extract the minimum bounding box that contain the detected face

Figure 4.4: Block diagram explaining proposed face detection system

4.4.1 Face Image

The face images are file of image entered to the system with true color (24-bit). The sizes of images are free, so we resize them into (256*256) pixels. Then the image is decomposing into its original three color bans (Red, Green, and Blue). As shown in figure (4.4.1).

Original image R


Figure 4.4.1: Face image and original three color bands

4.4.2 Color Transform

Many research studies that the chrominance components of the skin-tone color are independent of the luminance component (Sangwine & Horne, 1998).

However, in practice, the skin-tone color is nonlinearly dependent on luminance. The proposed detection system adopts the YCbCr space since it is perceptually uniform, is widely used in video compression standards (e.g., M PEG and JPEG), and it is similar to the TSL space in terms of the separation of luminance and chrominance as well as the compactness of the skin cluster. For this reason, in this stage RGB image would be converted to YCbCr color system using the equation (4.1)

Figure (4.4.2) explains color transformation

Y = 0.299 R + 0.587 G + 0.114 B………. (4.1.1)

Cb= (-0.169 R)+(-0.331 G)+ 0.5 B ………. (4.1.2)

Cr= 0.5 R +(-0.418 G) +(-0.081 B)………. (4.1.3)

Original image Y

Cb Cr

Figure 4.4.2: An example of YCbCr Transform space color system

4.4.3 Proposed Color-space Transformation

In the YC6Cr color space, chrominance (Cb and Cr) can be regarded as function of the luminance (Y), i.e. Cb (Y) and Cr (Y). Let the new transformed chrominance be C'b(Y) and C'r(Y). the skin color model is specified by the center (denoted as Cb(Y) and Cr(Y) and spread of the cluster (denoted as WCb(Y) and WCr(Y)) see Equations 4.2-4.5), and is used for computing the transformed chroma (Ibrahim, 2005).

If Y<Kl or Kh<Y

If YÑ” [Kl, Kh]

If Y< Kl

If Kh<Yhh

If Y<Kl

If Kh ≤ Y

If Y<Kl

If Kh ≤ Y

Where Ci in Equation (4.2) and (4.3) is either Cb or Cr, WCb=46.97, WLCb=23, WHCb=14, WCr=38.76, WLCr=10, Kl=50, and Kh=188.

These parameter values are estimated from training samples of skin patches from a subset of suggested image library. Y min and Y max values in the YCbCr color space are 16 and 235, respectively. Figure (4.4.3) show an example of this transform.

Original image Y

b r

Figure 4.4.3 : An example of YĆbĆr New Transform space color system

4.4.4 Skin color Tone Detection

As the color space transformed YC'bC'r in the previous section, that it was obtained then the skin color tone detected based on the C'r, when the value of C'r is larger than 1, then it indicate a skin color tone as shown in figure (4.4.4).Figure (4.4.5) shows the projection of face area.

Original image morphed mask morphed face

Figure 4.4.4: The skin color tone detection.

Original image bounded image

Figure 4.4.5: Projection of face areas

Image Resize


Original color

Face image









Detect Skin Color

Tone Region






Face Area


Figure 4.4.6: The Proposed face detection system stages