Kernel optimization based on salient region detection for image deblurring
Guanghui Xu, Xiangbo Xie, Kaixin Fan, Guojian Zheng · 2015
Camera shake commonly leads to image degradation during long time exposure. Recovering the degraded image is an ill-posed problem. The quality of a deblurred image is closely related to the correctness of the estimated blur kernel, and the incorrect kernel will lead to severe ringing artifacts after deconvolution. In this paper, we propose a blur kernel optimization method based on salient region detection of kernel image. Considering the inherent structure and sparse nature of the blur kernel, our method applies kernel-image signature to detect the trajectory of kernel and then extracts it from the sparse background. After building the finer kernel, we use total variant deconvolution algorithm to reconstruct the sharp image. Experiment results on synthesized and real-life images show that this approach is competitive with other state-of-the-art algorithms.