Reduction of Blocking Artifacts in Compressed Images using Non-Separable Fractional Fourier Transform

Abstract

Image compression has become an important aspect for many multimedia applications to fulfill the need of processing image for storage space, transmission bandwidth and representation with reduced cost. Image compression enables the autonomous machine to represent image utilizing less bits. Block-based compression algorithm with different transform techniques had been long used to compress image in a lossy manner. However, the reconstructed images from compression produces annoying blocking artifacts near the block boundaries, particularly in highly compressed images, as each block is transformed and quantized independently. Several techniques or algorithms have been proposed by researchers, both in spatial and frequency domains, for reduction of these blocking artifacts with varied degree of success. An image compression algorithm using Non-Separable Discrete Fractional Fourier Transform (NSDFrFT) employing any one of the suggested Bicubic interpolation and Nearest Neighbor interpolation apart from Bilinear interpolation as a transform technique has been proposed. The algorithm divide image into sub-images known as blocks, processing each independently. The NSDFrFT resulted in less MSE in blocked region than image compression using DFrFT and JPEG. However, if the different variations of NSDFrFT given as NSDFrFT-Nearest Neighbor Interpolation, NSDFrFT-Bilinear Interpolation and NSDFrFT-Bicubic Interpolation are compared then NSDFrFT-Bicubic Interpolation performs better than the other two variations. The performance of the transform techniques has been analyzed using various image quality metrics (IQM) among which GMSD is faster to calculate and provides high predictive accuracy. Thus, relying on accuracy of simulations, NSDFrFT results in structurally similar high subjective quality reconstructed image with reduced blocking for low frequency images at high compression percentages.