Disadvantages Of Earlier Image Retrieval Systems 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.

This paper suggests the system to recover images based upon dominant shades in the foreground area of the picture. This Paper researches the visualization recovery system dependent upon color . In order to access this information, an adequate recovery procedure is needed. The foreground of the visualization just gives semantics contrasted with the underpinning of the representation. This is the central theme of the proposed work.

[1] "Our world is dominated by lots of image informative content and a huge amount of such information is being added daily". It could be unrealistic to adapt to this eruption of superficial information, unless they are composed such that recovery of them effectively and successfully is conceivable.

Content based Image recovery is a swearing methodology to pursue through an image or a picture database by means of an image or picture features namely color, texture, shape, design, pattern or any mixed combinations of them. Color is an essential sign for a picture or image retrieval. Color isn't just adding delights to representations and yet gives more informative data, which is utilized as a compelling apparatus within substance based representation recovery. In Color Indexing, for any given Query picture the objective is to recover or retrieve every last trace of the representations whose color is comparable to the aforementioned of question picture.

2. Background

2.1. Existing System:

Inexpensive image capturing and various storage technologies have allowed tremendous collections of digital images. However, as [2] the database grows, the difficulty in order to find a relevant image increases. The most used traditional and common method of image retrieval is to utilize some method of adding metadata such as keywords, captions or descriptions to the images so that retrieval can be performed over the annotation words. [3] Manual image annotation is a time-consuming process. [4] It is laborious and expensive. To overcome this, Content Based Image Retrieval (CBIR) is presented, which points at keeping away from the utilization of text based portrayals and rather recovers pictures dependent upon their superficial similitude to a client-supplied inquiry picture. This recommended venture means to advance one such Content Based Image Retrieval System.

2.2. Proposed System

The proposed system is a stand-alone system. The user can maintain a database of images of numerous formats. He can add, delete, retrieve the images of his choice. For every image that is present in the database, the dominant color of its foreground object is stored. In order to retrieve images, the user should provide an enquiry or a query image to the system. With respect to this query image, the images which are similar will be displayed in such a way that the most similar image will be displayed first followed by the next similar image and so on. Different types of images are available in the database. So while adding images, we store them under different categories.


3.1 Disadvantages of Earlier Image Retrieval Systems

When introducing the major hypothesis of substance based image recovery, we will briefly look its development. Introductory work take on visualization recovery could be followed back to the late 1970s. [5] In 1979, there was a gathering that occurred in Florence, on Database Techniques for Pictorial Applications. On account of then, the provision potential of representation database administration strategies has lured the regard of countless analysts. Early procedures did not incorporate progressed headlines like surface emphasizes yet they were enabled on the printed annotation of images. [6] In different expressions, representations were first elucidated with substance then after that looked using a word-based system from traditional database management systems. Message-based visualization recovery utilizes time honored database methods to maintain postures. [7] By the utilization of content depictions, pictures might be ordered by topical or semantic pecking orders to expedite effortless access, searching and route dependent upon standard Boolean inquiries. Inasmuch as, mechanically created clear messages for a wide range of pictures is not practical, so the majority of the words-based picture recovery frameworks need manual annotation of representations. Clarifying visualizations manually is an exorbitant and bulky undertaking for gigantic picture databases, and is frequently, connection-delicate, subjective and fragmented. Subsequently, it is troublesome for the customary message-based routines to uphold an assortment of work-ward inquiries. Due to the above mentioned disadvantage there arose a situation where there is much need to implement a new approach. Thus the new approach that came into power is Content Based Image Retrieval.

3.2 Content Based Image Retrieval

Content based image recovery, a method which utilizes the visual substance to venture representations from vast scale image databases according to users' interest, has been an active and quick progressing research area in 1990s. Throughout the past decade, striking advancement has been made in both theoretical research and system development. However there remain numerous challenging research situations that continues to attract various developers and researchers from different disciplines.

They are certain basic approaches for Content Based Image Retrieval are :

3.2.1 Using Histogram-based Matching.

3.2.2 Using Dominant Color based Indexing.

3.2.1 Using Histogram-based Matching

[8] The histogram-based strategy is truly suitable for color visualization recovery since they are invariant to geometrical informative content in representations, for example interpretation and pivot. Histogram crossing point technique is to measure the crossing point territory between two pictures' histograms. They are more often than not named as reference visualization (R) for the inquiry enter and model representations (M) from the representation database. Similitude measure between R and M is then performed by computing the histogram convergence I (R, M). The heftier the worth I (R, M) the more comparative the pictures R and M are. Representations M might then be stacked up from the representation database. The same shade circulation histograms between distinctive shine states of the two advanced representations consequence in littler convergence esteem and make the profoundly on the surface comparable representations ends up being easier stacked up.

3.2.2 Using Dominant Color based Indexing

[9] Dominant color region that is present in an image can be represented as a connected fragment of homogeneous color pixels which is observed by human vision. Image Indexing is dependent upon this notion of predominant color present in the display picture. The sectioned out dominant regions along with their additional features are used as a source for retrieving similar images from the database. The path of the image, number of regions recognized, region information like normalized range, color and area of each particular region are stored in the file for future reference processing..

They are some drawbacks with the above mentioned approaches. Histogram approach is simple and easy but it does not allow for geometric information about images i.e. rotation and translation. In Dominant color locale based indexing the fundamental burden is that it never recovers the same objects of shifting sizes as the comparative visualization. For the more modest question background can be the predominant locale as demonstrated in figure1. 1, while in greater protest that question it is predominant. Granted the semantics of the items are same, they are not recovered as comparative visualizations.

Fig 2.1. Images with different Foreground Object sizes

The recommended strategy can answer this situation on account of thinking about just the foreground are a qualified data and ignoring background portions.

Proposed Method

3.3 Dominant Color Identification dependent upon Foreground Objects

Dominant color identification of foreground objects retrieves the number of similar images based on the foreground color irrespective of size and background color. The foreground information on the images is enough to identify the images properly. This is implemented by the proposed algorithm and by using some concepts like

1. Image Segmentation.

2. Color Space Catergorization.

3. Color Matching and Dominant Color Identification.

4. Retrieval method.

3.3.1 Segmentation of the Image.

[10] Image segmentation is the motivation of this research work, and is used to distinguish this technique from previous works of image retrieval based on dominant color Identification. The color image is converted into the grayscale image and then using the threshold method that will be converted into the binary image. In binary image the foreground is represented by the maximum value of intensity (1) and background is represented by the minimum value of Intensity (0). The binary image is converted into the color image by retaining the color values only in the foreground of the image.

3.3.2 Color Space Catergorization.

The whole RGB shade space is depicted utilizing color classes. This is condensed into a shade look-up table. A more diminutive set is more advantageous in light of the fact that it gives a coarser portrayal of the shade of the area along these lines permitting it to stay same for certain fluctuations in imaging conditions. [11] The color lookup table in Table 1, consists of 25 shades looked over 256 shade palette table. The proficiency of recovery framework might be upgraded whenever the prevailing color might be recognized within the more minor set of shades. Whenever the whole RGB color space is utilized for recognizing the predominant shade of a picture then the proficiency of the recovery technique can be diminished.


Black 1 1 1

Sea green 1 182 1

Light green 1 255 170

Olive Green 36 73 1

Aqua 36 146 170

Bright Green 36 255 1

Blue 73 36 170

Green 73 146 1

Turquoise 73 219 170

Dark Red 109 36 1

Blue Gray 109 109 170

Lime 109 219 1

Lavender 146 1 170

Plum 146 109 1

Teal 146 182 170

Brown 182 1 1

Magenta 182 73 170

Yellow Green 182 182 1

Floral Green 182 255 170

Red 219 73 1

Rose 219 146 170

Yellow 219 255 1

Pink 255 36 170

Orange 255 146 1

White 255 255 255

Table 1. Color Lookup table.

3.3.3 Color Matching and Dominant Color Identification

The image that is segmented is modified into smaller color set combination image. It involves mapping all pixels to their categories in color space. For each pixel in the image, a color is selected from a smaller set of predefined colors which are very near to image pixel color value it will be stored as new color pixel in the image.

The dominant color of the foreground area representation is resolved by the color reaction of every pixel in the changed visualization and archived in the color lookup table. This table is sorted in sliding request and after that the first event color could be the predominant color of the closer view of the separate representation.

3.3.4 Retrieval Method

[10] The suggested strategy Dominant shade distinguishing proof of forefront questions is utilized for every last trace of the color representations that are put forth in the database. It is utilized to figure out the forefront predominant color and after that the concentrated headline of prevailing shade and are saved. Whenever the inquiry picture is given by the client , the overwhelming color for frontal area informative data is recognized and retrieved. This recovery method distinguishes the database representations whose frontal area overwhelming color is compared to the closer view prevailing shade of inquiry visualization. Those pictures are recovered as the comparable pictures for the questionable picture

A brief overview of the proposed project system architecture .

Fig1 : Depicts the architecture for the proposed system

From the above figure the proposed project is presented in a format in which it can do to develop during the early research phase and also in system architecture phase.

4. Design architecture.

We can divide the system architecture in to various algorithmic parts in order to implement the proposed system

1. Identify the query image

2. Convert the query image into a Gray scale image.

3. Finding the threshold value of the Gray scaled query image.

4. Finding the Background intensity on the query image.

5. Finding a binary value for the query image.

6. Find the dominant color.

Many of the above mentioned steps are implemented by using algorithms.

4.1 Identify the query image

Initially we need to consider a color query image for which we need to identify the search results. This image is loaded and then we retrieve relatively similar images from the database using various techniques that use the dominant color identification of foreground objects.

4.2 Covert the query image into Gray scale image.

The color query image is taken and by using the concept of the gray scale algorithm we convert the color query image to gray scale image. After completing the process an intensity histogram will be developed or constructed.

Purpose: To convert the color image into a grayscale image and construct an intensity histogram

Input : CI, color image

Output : GI, grayscale image

IH, intensity histogram, an array of size 256


Step1. Collect the green, red and blue values of a pixel.

Step2. By using mathematical application convert those numbers into single

Gray value.

Step3. Replace, the original green, red, blue values with the newly acquired gray


Step 4. Focus is given mainly to step 2, using those values we calculate using the


Gray = ( Red + Green + Blu) / 3

4.3 Finding the threshold value of the Gray scaled query image.

This can be implemented by using Threshold value algorithm. In order to find the intensity threshold value based on which the image will be separated into foreground and background regions, we need to consider the output of gray scale image i.e. intensity histogram as an input in order to calculate the threshold value.

Purpose: To find the intensity threshold value based on which the image will be separated into foreground and background regions.

Input : IH, intensity histogram

Output : T, threshold value

Method :

Step 1. Least intensity is incremented by 1 and the if the value of I is less than highest intensity we increment the I value every time by 1.

Step 2. Calculate Nb, the percentage of pixels with intensity, Y< I

Step 3. Calculate MB, the weighted mean of pixel intensities where you<I

Step 4. Calculate No, the percentage of pixels with intensity, Y> I

Step 5. Calculate Mo, the weighted mean of pixel intensities where you>I

Step 6. Class variance, var=Nb*No* ( MB- Mo) 2

Step 7. If the value of max_var is less than four then,

Step 8. max_var=var; and value of T=i.

4.4 Finding the Background intensity on the query image.

The background intensity can be achieved by using background algorithm. The gray scale image and the threshold will be taken as the inputs and the desired background intensity is generated.

Purpose: To find the intensity that represents background

Input : GI, Grayscale image T, Threshold

Output : BI, an intensity that represents background

Method :

The TLI=intensity value of top-left pixel of GI.

TRI=intensity value of top-right pixel of GI

The BLI=intensity value of bottom-left pixel of GI

BRI=intensity value of bottom-right pixel of GI


4.5 Finding a binary value for the query image.

This can be achieved by using a binary algorithm, after forming the binary image. In order to produce a binary image the binary algorithm uses grayscale image, threshold and background intensity as an input. This image will be presented in black and white format. The white portion generally represents foreground and the black portion represents background.

Purpose: To form a binary image where the white portion represents foreground and black

Portion represents background

Input : GI, Grayscale image

T, Threshold

BI, Background intensity

Output : BImg, Binary image


Step 1. For each pixel of the grayscale image, GI.

Step 2. Find intensity value , Y value of the pixel

Step 3. If the value of Y is less than the T value. Then,

Step 4. If the value of BI is less than the T value. Then,

Step 5. The value Y=0 then the pixel value belongs to the background. Else,

Step 6. The value of Y=1 then the pixel value belongs to foreground. Else,

Step 7. If the value of BI is less than T value. Then. Else,

Step 8. If the value of Y=1 then the pixel belongs to foreground. Else,

Step 9. If the value of Y=0 then the pixel belongs to the background.

4. 6 Find the dominant color.

Finally we will be able to find out the dominant color of the proposed system after passing through the various above mentioned stages.

5. Design Scenario

The design of the proposed system depicts the software in various ways. First, the architecture of the system or product must be presented. Then, the various interfaces that connect the software to end-users or to other systems and devices. The programming parts that are utilized to build the suggested framework are at first planned. The design view of the proposed project may represent a different design or action, the final goal is to bind up to the set of basic design concepts that guide the later part of the proposed project i.e., coding and further implementation.

5.1 Scenarios

Various scenarios of the proposed system are given below. Every scenario which is listed below may result in a successful or a failure operation of the system. Every scenario will be dealt with very carefully while developing the proposed system. The various scenarios are:

Adding an image to the database

Separating foreground of the image from its background.

Finding a dominant color in the foreground of the image.

Deleting images from the database.

Retrieving Images from the database whose dominant color in the foreground object matches that of the query image supplied by the user.

1. Add Image

2. Delete Images

3. Retrieve Images

The description of each functional requirement is as follows:

5.1.1 Add Image

The user can add one image at a time to the database. If the image is not a compatible format, then the system should report an error displaying that the image format is not supported. Also, all the compatible image formats should be displayed. There is no constraint on the size of the image. While adding images to the database, the category to which it belongs should be specified.

5.1.2 Delete Images

The user can delete any number of images at a time from the database. The user has to specify the category of images from which he is intended to delete some images. If there are no images under that category to be deleted, then the system should report an error displaying that no more images are available.

5.1.3 Retrieve Images

The user has to supply a query image to the system. If the query image is not a compatible format, then the system should report an error displaying that the image format is not supported. Also, all the compatible image formats should be displayed. The user should also specify the image category from which he is intended to retrieve the images. Then, every image which is already in the database under that category and whose dominant color of its foreground object is almost similar to that of the query image is retrieved and displayed for the user.

6. Conclusion and Furutre scope

The suggested procedure of dominant color identification dependent upon frontal area questions is a considerable system to recover the representations dependent upon color. We choose foreground objects as a source to retrieve color because it has semantics compared to background thereby can retrieve images more efficiently. We are doing this retrieval based on color of the image. Further this can be extended by using shape, size.