Color Image Retrieval Based On Edge Detection And Adaptive Segmentation

With the rapid development of network technology and the increasing popularity of the image acquisition device, the source of the digital image is expanding constantly, at the same time, both the amount and the type of the image data is becoming more and more. In order to find the needed information from these massive image data quickly and efficiently, image processing technology need to develop rapidly. Image retrieval technology has become a hot spot of people’s research, and especially the content-based image retrieval technology is widely concerned. Image segmentation is the key step between image processing and image analysis.A key step of CBIR is feature extraction, and the foundation is the effective extraction of low-level visual features, which is the most studied part by researchers currently. Low-level visual features mainly include color features, texture features, shape features and so on. How to extract these features from images effectively is the most important question, and it is also a difficult problem. In the CBIR system there exist a semantic gap between low-level visual features and high-level semantic information. The region-based image retrieval is one of the effective strategies to solve this problem. First use image segmentation to divide an image into a number of specific regions which have unique properties, and extract the interested target at the same time, then complete image retrieval with region matching. Based on this, this paper does the following works:1. An image retrieval algorithm, the inertial product energy-based image retrieval algorithm, is proposed. When extracting edges in the noise image, the noise is often amplified. To improve this problem, in this paper, the inertial product energy-based edge detector is used in image retrieval. Compared with the Canny operator, the inertial product energy-based edge detector can extract the edges effectively, and is much better in suppressing noise. Based on the inertial product energy-based edge detector, use the Fourier descriptors and the color edge histogram as image features in image retrieval. The experimental results show that this method can accurately retrieve image, and the most important is to improve the effect of image retrieval for noise images which is hard to deal with.2. Propose a region-based color image retrieval method, a combination of adaptive segmentation and Multi-region matching. This paper present an adaptive image segmentation method first, segment images with MS first, and then use the Ncut method merge the over-segmented regions get from the first step. It is worth noting that here we do not need to pre-set the partition number in order to finish the merging process, but according to the features of the image which is an adaptive strategy. On the basis of image segmentation, in order to make full use of the information of image regions, introduce multi-region matching to complete the similarity matching between images. The experimental results show that the proposed region-based image retrieval method gets an ideal effect in the recall, precision and sorting.