UBC Theses and Dissertations
Colour image coding using linear transformation and block quantization Seecheran, Cris Anand
A bandwidth compression and redundancy reduction scheme was investigated in a digital colour image transmission system employing a linear transformation and block quantization technique. Colour images were represented by the N.T.S.C.'s Red, Green, Blue and Y, I, Q Receiver Primary and Transmission Co-ordinate systems respectively. The Y, I and Q signals were assumed to constitute a homogeneous Gauss-Markov random field modelled by an exponential correlation function of the form R(x,x¹,y,y¹) = exp[-α |x-x¹| - β|y-y¹|], where α, β are the correlation coefficients in the horizontal and vertical directions, respectively. Linear transformations employed were the discrete Fourier and Hadamard transformations. In the block coding scheme, blocksizes of 8 x 8 and 16 x 16 picture elements were used and in the quantization strategy, both optimum uniform and optimum non-uniform quantizers were considered. The Fourier and Hadamard coding schemes were evaluated in terms of the mean square error between original and reconstructed images, and by subjective preference. Based on the theoretical results, the Hadamard transform was superior to the Fourier transform for bit assignments below approximately 1.0 bits per picture element (bits/pel.), on each of the Y, I and Q signal planes. The experimental results showed the Hadamard coded images to be superior to the Fourier coded images from both the mean square error criterion and from a subjective evaluation of the processed images. Pictures coded with blocksizes of 16 x 16 picture elements were superior to those coded with 8x8 blocksizes. There was no noticeable difference in image quality between the optimum uniform or optimum nonuniform quantizer. Good quality images were obtained for an assignment of 2.0 bits/pel. on the Y signal and 0.375 bits/pel. on the I and Q signals. For some Hadamard transformed images, good quality reproductions were obtained for a total bit assignment as low as 1.75 bits/pel. Some experiments employing simple zonal low pass filtering schemes were also carried out when transform coding the entire image in its 256 x 256 format. When using this scheme, as compared to the block quantization scheme, better quality images were obtained for the Fourier system for equivalent bit assignments. There was not much difference for the Hadamard coded images.
Item Citations and Data