Research Of Image Retrieval Based On Pribalility Texture And Color

With the emergence of massive images in the Internet and multimedia database, how to browse and retrieve images quickly becomes an important problem to be solved. Image contains text information and content information. Thereinto, text information is the literal description of the image content, and in text-based image retrieval, manual annotation needs great workload and machine annotation is with low accuracy. However, content information can really reflect the content of images, and content-based image retrieval (CBIR) doesn’t have subjective effects caused by manual annotation. Therefore, content-based image retrieval becomes one of the most interesting technologies in image retrieval field.At present, CBIR researches are mainly based on low-level vision features, semantic, automatic annotation, relevance feedback and fast retrieval. Based on low-level vision features, this dissertation uses image multi-scale analysis and related techniques in the color area to make a research on CBIR. The major work of this dissertation can be summarized as following: (1) With the depth analysis of probabilistic statistic of image retrieval framework and related technology, we compare the performance of probabilistic statistic models and selects the appropriate model which can speed up the retrieval efficiency. In addition, we use two general parameter estimation method to estimate the parameters of these statistic models.(2) A RGB-based multi-scale analysis of image retrieval algorithm is proposed based on a future study on multi-scale analysis of CBIR. The initial multi-scale analysis of retrieval algorithm lack of important color information, we make use of RGB color space and its characteristics in the image preprocessing stage which can retain the color information in the image multi-scale analysis. The experimental results show that the proposed algorithm is superior to the traditional algorithm based on multi-scale analysis. Thereinto, corresponding to the proposed algorithm, average retrieval rate and average normalized modified retrieval rank have been raised by3.65%and25.8%, respectively.(3) A novel multi-feature algorithm is proposed based on probability statistics framework. In order to overcome the shortcomings of lack of color information and improve probability statistics retrieval performance, we combine with the global dominant color and probability statistics. In the improved experimental scheme, combinative features including texture and color are utilized for second retrieval after linear weighting. The experimental results on the2600images database obtain superior performance of our proposed method compared with the single feature image retrieval and the probability statistics retrieval, recall rates and average normalized modified retrieval rank have been raised by about30%and22%, respectively.