2017 IEEE Second International Conference on Data Science in Cyberspace (DSC) (2017)
Shenzhen, Guangdong, China
June 26, 2017 to June 29, 2017
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/DSC.2017.56
Extracting image features effectively and efficiently is an important issue in the area of image processing and matching. Most existing methods deal with images in the pixel domain while almost all digital images such as JPEG images are stored and processed in compressed domain where discrete cosine transform (DCT) -based method is used as one of the most popular data compression techniques. In this paper, we propose a method to compute difference of Gaussian space (DOG) and estimate SIFT feature descriptors in compressed domain using image pyramids. The main steps of our scheme include fast construction of DOG scale space in DCT domain, feature blocks extraction (replacement for feature points) and feature descriptors computation using image feature pyramids. Experiments show that our method can retain one half or more feature points and feature matches compared to original SIFT algorithm while time consumption is much less, which means that it can accelerate the procedure of JPEG image processing.
Feature extraction, Discrete cosine transforms, Dogs, Transform coding, Standards, Image coding, Histograms
C. Fei, B. Liu and N. Yu, "Extracting Sift Keypoints in DCT Domain," 2017 IEEE Second International Conference on Data Science in Cyberspace (DSC), Shenzhen, Guangdong, China, 2017, pp. 98-101.