Research On New Image Compression And Optimization Algorithms

Image compression is employed to reduce redundancy of the raw image data so as to save lots of storage space or transmission bandwidth.Image down-sampling is utilized very frequently in image compression and it can be implemented by reducing the image resolution or removing the transform coefficients.Adopting down-sampling in image compression saves bit-cost,and thus potentially improves the compression efficiency.In this thesis,we propose a novel transform down-sampling method and a novel error compensation method for the compression of image signals.The former method is proposed aiming at low down-sampling distortion and the later one is designed targeting on a low color distortion.Firstly,we propose a novel reconstruction-guided transform down-sampling algorithm for image compression.Compared with the traditional method,our proposed method produces the down-sized coefficient block with a low reconstruction distortion,and this coefficient block is formed as sparse as possible.With our proposed downsampling algorithm,we can compress image not only with a low bit-cost but also with a high reconstruction quality,which finally achieves a high compression efficiency.Then,we propose a novel error compensation algorithm for the compression of color images.In the traditional color image compression,the source RGB image must be converted to the YUV image and then compressed with a low YUV distortion.After compression,the YUV image is converted back to the RGB space for output.Such an RGBto-YUV based compression cannot guarantee a high quality for the compressed RGB image.To tackle this problem,we propose an error compensation algorithm for the color image compression.Specifically,the down-sampling errors and the compression errors occurred on the chrominance components(U and V)are coupled together and added to the luminance component(Y).With this compensation,we can achieve a low distortion and a high efficiency for the compression of color images.Experimental results demonstrate that applying our proposed methods to the practical image compression can achieve a higher compression efficiency compared with the traditional methods.