Learning In-Place Residual Homogeneity for Image Detail Enhancement

In this paper, we put forward and demonstrate a novel method in image and video detail enhancement-- in-place residual homogeneity (IP). In-place residual homogeneity is a regular law we find in testing different blocks in database, that is, residual blocks with slight different resolutions hold homogenous structures. By learning this homogeneity, we guess that it might be a good description of image's detail layer. Then images are enhanced by designed framework and accelerated by proposed fast in-place search method. Unlike most algorithms that need to adjust parameters by manual to get best performance, our approach is adaptive. Besides, many algorithms will change images' intensity, but our IP can keep natural images from over enhancement. Moreover, IP is also robust to low bit rate H.265 encoder and decoder system and runs faster than most popular methods. The last but not least, it can be easily FPGA implemented as well. Numbers of experiments testify that our algorithm is robust with good performance both subjectively and objectively.