Study On Key Techniques Of Region-based Image Matching Algorithm

The key techniques on region-based image retrieval and matching algorithm is elucidated in detail in this paper. Several key phases in low-feature-based image retrieval are feature extraction, image segmentation, region representation and image matching. An image is firstly partitioned into blocks with 4×4 pixels, extracting a feature vector for each block. The feature vector is composed of six features; those are 3 color features and 3 texture features. Such integrating feature vector is used for building K dimension gaussian model, whose parameters are estimated by an expectation-maximization (EM) algorithm, and then the resulting block-cluster memberships provide a segmentation of the image. For the purpose to lower the inaccuracy caused by the image segmentation, this paper uses a fuzzy feature representation method based on the Cauchy function to represent the region features. In the phase of region matching, this paper elucidates some matching algorithms commonly used, especially IRM and UFM matching scheme. In order to improve the retrieval efficiency, for several categories of images which have large background areas and simple color and texture, this paper makes some improvements based on the IRM method. Because objects and background in the images are separated after image segmentation and the objects occupy smaller areas than the background, the smaller areas label as objects, the rest as background. Since the UFM is faster than the IRM and the IRM is, to some extent, better than the UFM in image retrieval, the UFM is used in background region matching and the IRM is applied to object regions matching in this paper. To some extent, the experimental results show that the retrieval efficiency is improved apparently.