Contiguous Loss for Motion-Based, Non-Aligned Image Deblurring

In general dynamic scenes, blurring is the result of motion multiple objects, camera shaking or scene depth variations. As an inverse process, deblurring extracts a sharp video sequence from information contained in one single blurry image—it itself ill-posed computer vision problem. To reconstruct these frames, traditional methods aim to build several convolutional neural networks (CNN) generate different resulting expensive computation. vanquish this problem, innovative framework which can frames based on CNN model proposed. The motion-based image put into our and spatio-temporal encoded via pooling layers, output frames. Moreover, does not have one-to-one correspondence with any sequence, since sequences create similar images, so neither pixel2pixel nor perceptual loss suitable for focusing non-aligned data. alleviate problem novel contiguous function proposed focuses measuring Experimental results show that combined efficiently perform better than state-of-the-art methods.