Progressive edge-sensing dynamic scene deblurring

Abstract Deblurring images of dynamic scenes is a challenging task because blurring occurs due to combination many factors. In recent years, the use multi-scale pyramid methods recover high-resolution sharp has been extensively studied. We have made improvements lack detail recovery in cascade structure through network using progressive integration data streams. Our new and edge feature perception design deals with changes at different spatial scales enhances sensitivity blurred edges. The coarse-to-fine architecture restores image structure, first performing global adjustments, then local refinement. this way, not only correlation considered, but also residual information used significantly improve restoration enhance texture details. Experimental results show quantitative qualitative over existing methods.