Image Compression with Neural Network

Abstract

With many techniques available for image compression, the challenge to find the most efficient technique among them still prevails. Digitized images have replaced analog images as digital photographs in many different fields. In their unrefined form, digital images have need of a remarkable memory capacity for storage and large amount of bandwidth for transmission. In the last few decades, many researchers have been devoted to develop new techniques for image compression. More recently, wavelets have become a cutting edge technology for compressing the images by extracting only the visible elements. Our work presents implements the decomposition based Wavelet technique with it types such as Coiflets filter, Symlet filters and Daubechies filter and also neural network based Gradient technique been implemented. Also, a non-uniform threshold technique based on average intensity values of pixels in each sub band has been proposed to remove the insignificant wavelet coefficients in the transformed image. Experimental results are obtained to compare the Neural network based Gradient approach better to compress the image. But by fortunally, neural network technique does not work with 3-D images i.e color images. Here wavelet based technique, compactly supported (Daubechies) is tested on various images using two important performance parameters – MSE and PSNR results in best as compared to other coiflet and Symlet for compression of images in 3-D as well as in 2-D.
Key Word: Coiflet, Symlet, Daubechies, Gradient, Wavelet, Compression.