Image Resolution Enhancement of Single Color Image

Abstract

Digital Image Processing is an actively growing field which has many growing applications in Science and Engineering. The need of digital image processing is due to the improvement of pictorial information for human interpretations. Image processing may be classified into following categories: restoration, segmentation, and enhancement.Image Super-Resolution is the active field of image processing. Images and videos with high resolution are used in various fields like medicine, agriculture, pattern recognition etc. There are many applications for increasing the resolution of image but they are not very effective as they add physical artifacts such as noise or blur. There are various algorithms of image super resolution. Some algorithms do not consider the edges as the edges are most sensitive part of image. Some methods mainly focus on the edges leaving behind other part of image as it is. In this research work, we propose a method to increase the resolution of a color image. This framework involves various steps. First step involves the interpolation of image. The interpolation is done for scaling of the image using Bicubic interpolation. Next step is feature extraction where we extract the information of high frequency pixels of image. The features are extracted from the interpolated image where features are extracted using Gradient and Laplace filters. Gradient filter act as edge detector whereas Laplace filter improves low frequency pixels. We get four images from feature extraction as output in feature extraction block, first two by applying Gradient filter horizontally and vertically. Next two, we get by applying Laplace filter vertically and horizontally respectively. Then, the Principal Component Analysis reduction is used to get the most efficient information of feature extraction. Next step is to divide the image into non-overlapping blocks so that sparse representation method is performed on each block. After that we use Sparse Representation algorithm to perform the Super Resolution. Sparse Representation method involves two dictionaries for performing Super Resolution. Dictionaries are trained using Singular Value Decomposition (k-SVD) algorithm. Dictionaries are trained from the database which contains 60 images. The database contains both high resolution and low resolution version of images. Sparse representation is done on each block. Final step is to reconstruct the image by merging the output of sparse representation and the output of bicubic interpolation. At last, we get high resolution image as an output. The performance evaluation of the proposed method is based on Peek Signal to Noise ratio (PSNR), Structure Similarity Index iv (SSIM) values. The PSNR and SSIM values of our proposed algorithm are better than the Bicubic interpolation, Lancoz interpolation, k-neighbor algorithm etc. Sparse representation has better performance than various interpolation methods. The average PSNR value of proposed method exceeds than that of sparse representation by 0.24 which makes it more suitable than contemporary methods.