Research And Implementation Of Content-based Image Retrieval Techniques

Content-based Image Retrieval (CBIR) is a kind of retrieval, which directly use content of an image for information search, it is becoming an important direction in multimedia retrieval field. In order to analysis the information included in an image, the CBIR system always utilizes the color, texture, shape and other low level image features, to establish the feature vectors as retrieval index. In present time, the main CBIR method is similarity search based on multi-dimensions feature vector of image.The main research direction of the paper is a retrieval method based on representative color feature. After a review to the evaluation of image retrieval, we elaborated the key technologies in the content-based image retrieval and presented a spatial color indexing approach using the dominant colors of multi-resolution partitions. At first we made partitions to the image in multi resolutions (here it mains the different size of partitions), on the grounds of human visual sensation and psychology, we convert the RGB color spaces to HIS color spaces which could be at one with human's visual feature. Then we compute and draw out the color which has the most pixels from every partitions (we called it dominant color) under HIS color spaces. It provided a kind of method to say the feature vector quantity for every partition feature, and then form the color feature space of the whole picture.Because feature vectors are all high linked, so, in the retrieval based on content, to calculate the distance will consume a large amount of time. In order to reduce the dimension in the feature vectors, we provided the quantizing algorithms of the color space in detail in this paper with the amount of calculation reduced. In the similar degree of tolerance of the picture, we adopt the method of the comprehensive tolerance under the same resolution ratio and different resolution ratios, and introduce different weights according to users' feedback in the course of measuring.Be based on the above algorithm, we develop the image retrieval system to prove its validity and exactness. The system realizes this algorithm with Visual C ++, we have done a large number of experiments and obtained the better experimental result.