A neural and interpolation method for wavelet transform based image compression

The characteristics of providing spatial and frequency information in the transform domain by wavelet transform, plays a crucial role in forming a significance map of coefficients that needs to be coded in the case of image compression. This paper presents a wavelet based image compression technique where transmission of spatial locations of wavelet decomposed lower subbands coefficients is eliminated. The spatial position information of significant coefficients of subbands of each level can be found by interpolation of higher level subbands to lower level at decoder side. This in turn, saves the data corresponding to positional information of significant coefficients, further giving an opportunity to enhance image quality at same compression level by including additional wavelet coefficients. The simulation results show that the combination of interpolation and vector quantization (VQ) using neural networks performs better than baseline JPEG.