Research And Implementation Of Content Based Image Retrieval

With the rapid development of information technology, people are exposed to more and more image information. The advances in technology bring a lot of conveniences to our lives, and mass image data also broaden our horizons. How to help the user to retrieve the target image quickly from the huge image database has been a key issue to be solved. This technique is also known as image retrieval technology, which mainly includes two research directions, text-based image retrieval and content-based image retrieval. The research direction of this thesis is content-based image retrieval technology. In the computer vision system, the content of the image include color, texture, shape and so on, these features are all low-level image features, and it also include the semantic level information which is the user-oriented feature, it refers to the abstract concept of the image and the space relationship between the objects in the image.The paper study the content-based image retrieval technology, especially focus on the extraction of various low-level features. The goal of this paper is to improve the performance of the algorithm.Because the traditional color histogram cannot reflect the spatial relationship of color, so the paper proposes an adaptive weighted on block retrieval algorithm. Firstly, the image is divided into overlapping region, then, we extract color histogram from each region; secondly, according to the extracted color feature vector, we calculate the similarity between the central region and the surrounding region, and assign the weight of each region by the similarity, the higher the similarity, the greater the weight, the same token, the lower the similarity the smaller weight; finally we weight the color feature vector of each region and retrieve. Experimental results show that the proposed algorithm doesn't increase the retrieval time apparently but has the higher retrieval accuracy than others.Combine the multi-scale analysis feature of discrete wavelet transform, gray level co-occurrence matrix, and histogram of oriented gradients feature which can describe the appearance and shape features of local target region, the paper propose a new algorithm that GLCM and HOG retrieval method based on discrete wavelet transform. First we process the image by discrete wavelet transform, extract texture feature of high-frequency image, and gray level co-occurrence matrix feature of wavelet sub bands, then use the harris corner to detect the points of interest on wavelet sub bands, extract HOG feature of local region around the interest points, finally, based on the global and local information fusion, we combine the feature of the GLCM based on DWT, layered texture feature, HOG feature. Experimental results show that the retrieval efficiency of the proposed algorithm is improved.At last, the paper design and implement a platform of content-based image retrieval with simple, intuitive interface and other features. The platform is used to test the algorithms which proposed in this paper.