Neuro-Curvelet Model for Efficient Image Compression Using Vector Quantization

摘要

In many multimedia applications, such as image storage and transmission image, compression plays a major role. The fundamental objective of image compression is to represent an image with least number of bits of an acceptable image quality. A technique based on second-generation curvelet transform and Back-Propagation Neural Network (BPNN) has been proposed. The image compression is accomplished by approximating curvelet coefficients using BPNN. By applying BPNN into compressing curvelet coefficients, we have proposed a new compression algorithm derived from characteristic of curvelet transform. Initially, the image is translated by fast discrete curvelet transform and then based on their statistical properties; different coding and quantization schemes are employed. Differential Pulse Code Modulation (DPCM) is employed to compress low-frequency band coefficients and BPNN is used to compress high-frequency band coefficients. Subsequently, vector quantization is performed on BPNN hidden layer coefficients, thereby resulting in a reconstructed image with less degradation at higher compression ratios. For a given bits per pixel (bpp), the Curvelet Transform with Back-Propagation Neural Network (BPNN) gives better performance in terms of Peak Signal-to-Noise Ratio (PSNR) and Computation Time (CT) when compared to Wavelet Transform with BPNN and JPEG.