On the Efficiency of the Differential Pulse Code Modulation in Image Coding and Compression and an Implementation on a Digital Signal Processor Based System

This paper describes tools for the theoretical and experimental study of several DPCM methods in image acquisition and compression, with applications in Intelligent Visual Sensor development. The paper also presents conclusions about DSP based implementations of the described algorithms. 1.Introduction DPCM ([1][2]) is a general method of compression useful in various fields of the signal processing technology. Even today, when other tools offer a larger compression ratio , simplicity of the DPCM is a major advantage that results in a cheap implementation and in a higher compression speed. The basis of DPCM compression is cuantisation and coding of the difference between the predicted sample and the real sample .The predicted sample is often a linear combination of the preceding samples.In most applications the prediction is limited at the closest preceding samples of the predicted pixel. Shortly, the algorithm can be described in the following formula: e(i,j)=a1*e(i-1,j)+a2*e(i,j-1)+a3*e(i-1,j-1)+ r(i,j) (1) where e(i,j)is the image sample for the i,j co-ordinates a1,a2,a3coefficients of the prediction r(i,j)the residual error The residual error has a lower entropy than the original image and can be thus represented with fewer bits. Choosing the coefficients is often the main problem of the algorithm, and can be done for error minimisation and /or to obtain a higher compression ratio. An other problem in the algorithm evaluation is to establish measures of fidelity. A widely used measure of reconstructed image fidelity, used also in this paper, for an NxN image size is the average mean-square error defined as: N E e i j e i j NxN ms j N