An Efficient Content-based Image Retrieval System Integrating Wavelet-based Image Sub-blocks with Dominant Colors and Texture Analysis
ABSTRACT
Multimedia information retrieval is a part of computer science and it is used for extracting semantic information from multimedia data sources such as image, audio, video and text. Automatic image annotation is called as automatic image tagging or automatic linguistic indexing. It is the process in which a computer system automatically designates metadata in the form of keywords or captioning to a digital image. This application is widely used in image retrieval systems to locate and organize images from database. In this paper we have proposed efficient content based image retrieval (CBIR) systems due to the availability of large image database. The image retrieval system is used to retrieve the images based on color and texture features. Firstly, the image is partition into equal sized non-overlapping tiles. For partitioning images we are applying methods like, Gray level co-occurrence matrix (GLCM), HSV color feature, dominant color descriptor (DCD), cumulative color histogram and discrete wavelet transform. An integrated matching scheme can be used to compare the query images and database images based on the Most Similar Highest Priority (MSHP). Using the sub-blocks of query image and the images in database, the adjacency matrix of a bipartite graph is formed.

Get Help With Your Essay
If you need assistance with writing your essay, our professional essay writing service is here to help!
Essay Writing Service

INTRODUCTION:
Automatic image annotation is known as automatic image tagging or automatic linguistic indexing. It is the process in which a computer system automatically designates metadata in the form of keywords or captioning to a digital image. This application is widely used in image retrieval systems to locate and organize images from database. This method can be considered as multi class image classification with a large number of classes. The advantage of automatic image annotation is that the queries that can be specified by the user. Content based image retrieval requires users to search by images based on the color and texture and also is used to find example queries. The traditional methods of image retrieval are used to retrieve annotated images from large image database manually and which is an expensive, laborious and time consuming in existence.
Animage retrieval system is a computer system for searching, browsing and retrieving images from a largecollectionofdigital images. Most common and traditional methods of image retrieval use some methods of adding metadata such as captioning or descriptions and keywords to the images so that the retrieval can be performed over the annotation words. Image searchis used to find images from database and a user will provide a query terms as image file/link, keywords or click on some image and the system will return images similar to that query image. The similarity matching is done by using the Meta tags, color distribution in images and region/shape attributes.

Image Meta Search: – searching the images based on associated metadata such as text, keywords.
Content-Based Image Retrieval (CBIR):- This is the main application of computer vision to retrieve the images from image database. The aim of CBIR is used to retrieve images based on the similarities in their contents such as color, texture and shape instead of textual descriptions and comparing a user-specified image features or user-supplied query image.
CBIR Engine List: – This is used to search images based on image visual contents as color, texture, and shape/object.
Image Collection Exploration: – It is used to find images using novel exploration paradigms.

Content Based Image Retrieval:
Content based image retrieval is known asquery by image content(QBIC) andcontent-based visual information retrieval(CBVIR) and it is the application ofcomputer vision techniques to retrieve the images from digital image database. This is the image retrieval problem of finding for images in large image database. Content-based image retrieval is to provide more accuracy as compared to traditionalconcept-based approaches.
Content-based is the search that analyzes the contents of the image instead of metadata such as keywords, tags, or descriptions associated with that image. The term “content” in this context means textures, shapes, colors or any other information about image can be derived from the image itself. CBIR is popular because of its searches are purely dependent on metadata, annotation quality and completeness. If the images are annotated manually by entering the metadata or keywords in a large database can be a time consuming and sometime it cannot be capture the keywords preferred to describe its images. The CBIR method overcomes with the concept based image annotation or textual based image annotation. This is done by automatically.
Content Based Image Retrieval Using Image Distance Measures:-
In this the image distance measure method is used to compare the two images such as a query image and an image from database. An image distance measure method is used to compare the matching of two images in various dimensions as color, shape, texture and others. Finally these matching results can be sorted based of the distance to the queried image.
Color
This is used to compute image distance measures based on color similarity. This is achieved by computing the color histogramfor each image and that is used to identify the proportion of each pixel within an image which is holding a specific values. Finally examine the images based on the colors, which contains most widely used techniques and it can be completed without consider to image size or orientation. It is used to segment color by spatial relationship and by region among several color region.
Texture
Textures are represented as texels and are then located into a number of sets based on a lot of textures and are detected in the images. These sets are used to define texture and also detect where the textures are located in images. Texture measures are used to define visual patterns in images. By using texture such as a two- dimensional gray level variation is to identify specific textures in an image is achieved. Using texture, the relative intensity of pairs of pixels is estimated such as contrast, regularity, coarseness and directionality.Identifying co-pixel variation patterns and grouping them with particular classes of textures like silky, orrough.

Different methods of classifying textures are:-

Co-occurrence matrix.
Laws texture energy.
Wavelet transforms.

LITERATURE SURVEY:
In this paper a multscale context dependent classification algorithm is developed for segmenting collection of images into four classes. They are background, photograph, text, and graph. Here, features are used for categorization based on the distribution patterns of wavelet coefficients in high frequency bands. The important attribute of this algorithm is multscale nature and is used to classifies an image at different resolutions adaptively and enabling accurate classification at class boundaries. The collected context information is used for improving classification accuracy. In this two features are defined for distinguishing local image types in image database according to the distribution patterns of wavelet coefficients rather than the moments of wavelet coefficients as features for classification. The first feature is defined for matching between the empirical distribution of wavelet coefficients in high frequency bands and the Laplacian distribution. The second feature is defined for measuring the wavelet coefficients in high frequency bands at a few discrete values. This algorithm was developed to calculate the feature efficiently. The multscale structure collects context information from low resolutions to high resolutions. Classification is done on large blocks at the starting resolution to avoid over-localization. Here, only the blocks with extreme features are classified to ensure that the blocks of mixed classes are left to be classified at higher resolutions and the unclassified blocks are divided into smaller blocks at the higher resolution. These smaller blocks are classified based on the context information achieved at the lower resolution. Finally simulations shows that the classification accuracy is significantly improved based on the context information. Multiscale algorithm is also provides both lower classification error rates and better visual results [1].
This paper proposed content based image retrieval technique that can be derived in a number of different domains as Medical Imaging, Data Mining, Weather forecasting, Education, Remote Sensing and Management of Earth Resources, Education. The content based image retrieval technique is used to annotate images automatically based on the features like color and texture known as WBCHIR (Wavelet Based Color Histogram Image Retrieval). Here, color and texture features are extracted using the color histogram and wavelet transformation and the mixture of these two features are strong to scaling and translation of objects in an image. In this, the proposed system i.e. CBIR has demonstrated a WANG image database containing 1000 general-purpose color images for a faster retrieval method. Here, the computational steps are effectively reduced based on the Wavelet transformation. The retrieval speed is increases by using the CBIR technique even though the time taken for retrieving images from 1000 of images in database is only a 5-6 minutes [2].
This paper presents content based image retrieval scheme for medical images. This is an efficient method of retrieving medical images based on the similarity of their visual contents. CBIR-MD system is used to facilitate doctors in retrieving related medical images from the image database to diagnose the disease efficiently. In this a CBIR system is proposed by which a query image is divided into identical sized sub-blocks and the feature extraction of each sub-block is conceded based on Haar wavelet and Fourier descriptor. Finally, matching the image process is provided using the Most Similar Highest Priority (MSHP) principle and by using the sub-blocks of query and target image, an adjacency matrix of bipartite graph partitioning (BGP) created [3].
In this paper a content based image retrieval (CBIR) system is proposed using the local and global color, texture, and shape features of selected image sub-blocks. These image sub-blocks are approximately identified by segmenting the image into small number of partitions of different patterns. Finding edge density and corner density in each image partition using edge thresholding, morphological dilation. The texture and color features of the identified regions are calculated using the histograms of the quantized HSV color space and Gray Level Co- occurrence Matrix (GLCM) and the combination of color and texture feature vector is evaluated for each region. The shape features are computed using the Edge Histogram Descriptor (EHD). The distance between the characteristics of the query image and target image is computed using the Euclidean distance measure. Finally the experimental results of this proposed method provides a improved retrieving result than retrieval using some of the existing methods [4].
An efficient content based image retrieval system plays an important role due to the availability of large image database. The Color-Texture and Dominant Color Based Image Retrieval System (CTDCIRS) is used to retrieve images based on the three features such as Dynamic Dominant Color (DDC), Motif Co-Occurrence Matrix (MCM) and Difference between Pixels of Scan Pattern (DBPSP). By using the fast color quantization algorithm, we can divide the image into eight partitions. From these eight partitions we obtained eight dominant colors. The texture of the image is obtained by using the MCM and DBPSP methods. MCM is derived based on the motif transformed image. It is related to color co-occurrence matrix (CCM) and it is the conventional pattern co-occurrence matrix and is used to calculate the possibility of the occurrence of same pixel color between each pixel and its nearby ones in each image, which is the attribute of the image. The drawback of MCM is used to capture the way of textures but not the difficulty of texture. To overcome this, we use DBPSP as texture feature. The combination of dominant color, MCM and DBPSP features are used in image retrieval system. This approach is efficient in retrieving the user interested images [5].
In this paper content based image retrieval approach is used. It consists of two features such as high level and low level features and these features includes color, texture and shape which are present in each image. By extracting these features we can retrieve the images from image database. To obtain better results, RGB space is converted into HSV space and YCbCr space is used for low level features. The low level features are to be used based upon the applications. Color feature in case of natural images and co-occurrence matrix in case of textured images yields better results [6].
OBJECTIVE:

To retrieve images more efficiently or accurately.
To improve the efficiency and accuracy by using the multi features for image retrieval (discrete wavelet transform).
Image classification and accuracy analysis.
Time saving.
Robustness.

METHODOLOGY:

Discrete Wavelet Transform.
Conversion to HSV Color Space.
Color Histogram Generation.
Dominant Color Descriptor.
Gray-level Co-occurrence Matrix (GLCM).

ARCHITECTURE:
 
This architecture consists of two phases:

Training phase
Testing phase

These two phases of the proposed system consists of many blocks like image database, image partitioning, wavelet transform of image sub-blocks, RGB to HSV, non uniform quantization, histogram generation, dominant color description, textual analysis, query feature, similarity matching, feature database, returned images.
In training phase, the input image is retrieved from image database and then the image is being partitioned into equal sized sub-blocks. Further, for each sub-block of the partitioned image, wavelet transform is being applied. Then the conversion from RGB to HSV taken place preceded with non uniform quantization, inputted to histogram generation block where a color histogram is generated for the sub-blocks of the image. Then the dominant color descriptors are extracted and texture analysis of each sub-block of the image is done. Finally the image features from the feature database and the input image features are compared for the similarity matching using MSHP principle. Then the matched image is being returned.
In testing phase, the processing steps are same as training phase, except the input image is given as the query image by the user not collected from the image database.
OUTCOMES:

It provides accurate image retrieving.
Comparative analysis and graph.
Provides better efficiency.

CONCLUSION:
To retrieve images from image database, we can use discrete wavelet transform method based on color and texture features. The color feature of the pixels in an image can be described using HSV, color histogram and DCD methods, similarly texture distribution can be described using GLCM method. By using these methods we can achieve accurate retrieval of images.
REFERENCES:
[1] Jia Li, Member, IEEE, and Robert M. Gray, Fellow, IEEE, “Context-Based Multiscale Classification of Document Images Using Wavelet Coefficient Distributions”, IEEE Transactions on Image Processing, Vol. 9, No. 9, September 2000.
[2] Manimala Singha and K.Hemachandran, “Content Based Image Retrieval using Color and Texture”, Signal & Image Processing: An International Journal (SIPIJ) Vol.3, No.1, February 2012.
[3] Ashish Oberoi Deepak Sharma Manpreet Singh, “CBIR-MD/BGP: CBIR-MD System based on Bipartite Graph Partitioning”, International Journal of Computer Applications (0975 – 8887) Volume 52– No.15, August 2012.
[4] E. R. Vimina and K. Poulose Jacob, “CBIR Using Local and Global Properties of Image Sub-blocks”, International Journal of Advanced Science and Technology Vol. 48, November, 2012.
[5] M.Babu Rao Dr. B.Prabhakara Rao Dr. A.Govardhan, “CTDCIRS: Content based Image Retrieval System based on Dominant Color and Texture Features”, International Journal of Computer Applications (0975 – 8887) Volume 18– No.6, March 2011.
[6] Gauri Deshpande, Megha Borse, “Image Retrieval with the use of Color and Texture Feature”, Gauri Deshpande et al, / (IJCSIT) International Journal of Computer Science and Information Technologies, Vol. 2 (3) , 2011, 1018-1021.
[7] Sherin M. Youssef, Saleh Mesbah, Yasmine M. Mahmoud, “An Efficient Content-based Image Retrieval System Integrating Wavelet-based Image Sub-blocks with Dominant Colors and Texture Analysis”, Information Science and Digital Content Technology (ICIDT), 2012 8th International Conference on Volume:3 .