Significant Region And Color Information-based SIFT Algorithm In Image Retrieval

The rapid growth of network technology and calculation performance have been able to meet the huge demand of the image processing fine, such as search engine based image, video identification, robots and other important application. In these applications, image matching and retrieval technology as the most basic work, has a very important significance. According to the actual project application, this paper researches the current applied most widely popular image matching algorithm, SIFT, SURF, PC A-SIFT and so on. According to the principle analysis of SIFT algorithm, the paper analysis the shortage of SIFT algorithm in practical application.In feature extraction, by marking founded features in experiment, paper found that there exist a large number of background pixels which are extracted as features. There background pixels can increase error matching pairs during the matching stage. Therefore, this paper studies common significant region detection method, and puts forward a new method to extract notable region detection algorithm based on pixel variance, which makes all image features stay in object area.Due to the lack of color information in SIFT algorithm, an extended rgb model is proposed in this paper. This model is used to extract color information in the neighborhood of the feature. Finally, a48-dimensional vector is produced to represent the local color information of the feature. This method can increase the Information integrity of the feature.In match stage, a new match strategy-combine absolute distance and relative distance is proposes as image similarity measure.Experiments show that all the improvements are effective. Improved SIFT algorithm maintain the accuracy as well as reducing retrieval time by1/5, and error matching rate by10%. In the large image retrieval system, the improved SIFT algorithm is more practical.