The Study On The Advanced Algorithm Based On PCA-SIFT For Image Retrieval

Currently, with the developing of computer technology and multi-media, computer vision is widely used. It is extremely popular in people’s daily life to use the applications of digital image information processing. The traditional image retrieval algorithm is usually based on image content, name or scale to describe the key information in such image, as a result of artificial operation‘s subjectivity and uncertainty which cannot achieve efficient image retrieval request. To further explore, researchers proposed image retrieval method based on the content of the image, this method is based on the image color distribution, structure and texture, effectively avoid the disadvantages of artificial operation, and improve the efficiency and accuracy of image feature extraction, which also plays a great role in image retrieval of practical industrial applications.The local image feature description is an important part of image retrieval. Nowadays, SIFT is widely used in local feature descriptor because of its strong robust to image rotation, scale and illumination. When processing the local feature descriptors, PCA algorithm can obtain a set of non-linear variants after orthogonal transformation. This algorithm can effectively extract the major components in feature space, to better describe the image features dataset.Although the use of the original high-dimensional descriptor can accurately do the image matching and retrieval, the high dimension has also led to a lot of storage space wasted and time consumed. In this paper, the SIFT algorithm is improved, by PCA, proposed a new image local feature descriptor with low dimensions. After a great deal of experiment, this paper demonstrate that this new descriptor can greatly maintain the performance of original SIFT descriptor, and efficiently improve the image matching and retrieval with time.The proposed algorithm improve SIFT descriptor in a different way, it has a great value in improving the efficiency of image matching and retrieval.This paper first introduces the algorithm’s research background, significance and the situation at home and abroad. It also introduces the existing image retrieval application examples, then scale invariant feature transform algorithm and principal component analysis algorithm is introduced, the basic principle of the above two algorithms are analyzed in detail. PCA- SIFT algorithm is also described with new method in experiment. It detailed analyzes the overall frame of this algorithm and the improvement of different scales in the frame. Finally, the experiments have been carried on to test the advanced algorithm, and the results are analyzed and summarized.