Neuro-curvelet-based image compression technique for noisy images

Abstract

The main aim of image compression is to represent an image with minimum number of bits, so that the storage requirement can be reduced, thereby increasing the transmission rate without losing significant features of the image. The compression ratio is affected by noise, as it degrades the correlation between pixels. During capture, processing or transmission of the image, noise can occur. The noise possibly can be independent of or dependent on image content. On lossy image compression algorithms, the effect of noise has been studied in this paper. In order to study the effect of noise, the original images act as a reference to the reconstructed images. The reconstructed images are compared with the original images in terms of PSNR. The proposed image encoder integrates the features of curvelet transform with both radial basis function neural network (RBFNN) and back-propagation neural network (BPNN) separately and results are presented for both the cases. The case studies which consider images with noise prove the superiority of the techniques in terms of highly acceptable PSNR values. The merits of the proposed technique are further exemplified by comparing the results with those of JPEG and JPEG 2000.