Wavelet-Neural Network Based Image Compression System for Colour Images

There are many images used by human being, such as medical, satellite, telescope, painting, and graphic or animation generated by computer images. In order to use these images practically, image compression method has an essential role for transmission and storage purposes. In this research, a wavelet based image compression technique is used. There are various wavelet filters available. The selection of filters has considerable impact on the compression performance. The filter which is suitable for one image may not be the best for another. The image characteristics are expected to be parameters that can be used to select the available wavelet filter.

The main objective of this research is to develop an automatic wavelet-based colour image compression system using neural network. The system should select the appropriate wavelet for the image compression based on the image features. In order to reach the main goal, this study observes the cause-effect relation of image features on the wavelet codec (compression-decompression) performance. The images are compressed by applying different families of wavelets. Statistical hypothesis testing by non parametric test is used to establish the cause-effect relation between image features and the wavelet codec performance measurements. The image features used are image gradient, namely image activity measurement (IAM) and spatial frequency (SF) values of each colour component.

This research is also carried out to select the most appropriate wavelet for colour image compression, based on certain image features using artificial neural network (ANN) as a tool. The IAM and SF values are used as the input; therefore, the wavelet filters are used as the output or target in the network training.

This research has asserted that there are the cause-effect relations between image features and the wavelet codec performance measurements. Furthermore, the study reveals that the parameters in this investigation can be used for the selection of appropriate wavelet filters. An automatic wavelet-based colour image compression system using neural network is developed. The system can give considerably good results.