An Improved Image Inpainting Algorithm Based on Normalized Convolution

Although normalized convolution (NC) algorithm can be used to repair discrete isolated data loss, the image inpainting performance for the continuous block data loss using NC isn't good.A filter of neighborhood of pixel is defined, and a new image inpainting algorithm based on NC is proposed.The repair order for the loss block data is from the block borders gradually spread to the center.Then, each pixel of the image block is assigned to a confidence, and each pixel is set a confidence value from high to low according to the pixel's repairing order.That is the last repaired pixel is set the lowest confidence value.Finally, the pixels to be repaired are interpolated according to the around pixels' value and confidence.Experimental results show that the proposed algorithm is a practical image inpainting algorithm, and good performance can be obtained for the continuous data loss image inpainting.