Feature-Based Facial Image Coding Method Using Wavelet Transform

Chuxi Yang, Yan Zhao, Shigang Wang, Yuanyuan Liu

Abstract: This paper proposed a novel image coding scheme for facial images in the transform domain based on Scale Invariant Feature Transform (SIFT). With the characteristic of rotation scale invariability, SIFT descriptors are perfect to describe the details in an image. Our proposed coding scheme combine wavelet transform with SIFT algorithm. High frequency sub band images of the input image are encoded according to SIFT key points rather than pixel value on the encoder side. On the decoder side, we can get restoring patch from an image database according to key point matching. After clustering and transformation, the matching patches are used to restore the high frequency sub band images. Final reconstructed image was produced by wavelet inverse transform from low frequency sub band image and restored high frequency sub band images. Experimental results show that compared with HEVC and JPEG2000, the proposed algorithm can reach a higher compression ratio and a satisfactory reconstruction performance.