An edge preserving codebook generation algorithm in vector quatization for image coding is proposed and verified. Vector quantization technique has been known as a respective method for data compression at low rate but the reconstructed image suffers from staircase-like artifacts at sharp edges.
To preserve the edge integrity, a method based on HVS (human visual system) was proposed which is known as CVQ (classified vector quantization). The performance of CVQ technique is highly dependent on the classifier's resolvability with which the input vectors are classified with respect to the human visual system's response. However this technique is restricted to small size blocks such as 4 × 4, or 5 × 5 blocks and the misclassification of input blocks is an inevitable problem.
In this paper, a new edge preserving codebook design algorithm is suggested, which is an advanced form of LBG algorithm and does not use any classifier. The method inspects some frequency components of input blocks through discrete cosine transform and evaluates a measure which represents the amount of edge components. This measure is utilized as a weighting factor of input vectors in codebook design procedure and this fact can be regarded as if the probability density of input blocks are empasized on edge blocks. Through some simulations, the validity of the proposed method is verified.