Study On Image Search Engine Based On Color Feature Extraction

With the development of network technology, multimedia technology and the digitaltechnology, image resources are increasingly rich in the network. Various kinds of imagesearch engine based on Web spring up in order to satisfy the user’s retrieval requirement.Usually, the most thing that people concerns about is whether the search results can meettheir request. And the accuracy of the results depends on quality of the image matchingalgorithm.At first, beginning with the study of the quantization algorithm, the descriptionalgorithm, and the matching algorithm, based on color feature. Then, aim at the shortagethat some existing algorithms only concerns about the color rate but ignore the spacedistribution information of color in image, a new matching algorithm is raised, which isbased on the similar comparison of irregular shapes. The innovations of this algorithm areas follows.Firstly, the improving matching algorithm includes the space information, and makesthe retrieval results more accurate. The common expression methods of color feature,including color histogram method, the main color method and etc, usually concern aboutthe total ratio of color in image, without the consideration of color space information. Thus,this causes a problem that different images, which have the same color histogram, mayhave different contents because of the different color distribution.Secondly, the raised algorithm has an idea of blocking automatically, and it canseparate the image into different blocks according to the different objects in the image.Then, we can comparison the image block by block. This algorithm increases the accuracyof image search. Some improved image matching algorithm, such as the color histogrambased on blocks and the main color method based on regions, consider about the colorspace information. But, for this kind of methods, the process of dealing with imagesdepends on the fixed number of blocks, which has nothing to do with the characteristic ofactual objects in images. Once the number is set, the mistake might appear becausedifferent images have different contents and they might have a big difference with thesetting block number.Finally, the performance comparison between the existing algorithms and improvingalgorithm is studied. Then, a system is designed realized. Through the comparison of practical retrieval results, the evaluation of the improving algorithm is given. Experimentproves that the new algorithm has no relationship with the change of image, such as therotation, the translation and the size of image. It has an excellent stability, and can searchthe images that satisfy user’s requirement exactly.