Image compression of facial photographs based on BTC/TSVQ local processing

This paper investigates the performance of an image compression algorithm tailored to compress gray-scale facial photographs. The proposed technique treats blocks in the head region differently and adapts itself to the local nature of the face region. The proposed technique first locates the head-shoulder region and then locally processes block in the head region and encode active blocks containing edges with the Block Truncation Coding (BTC) [2,4,9,15] technique. It encodes inactive blocks having low intensity variations between its pixels with the Tree Structured Vector Quantization (TSVQ) [13] technique. The cheeks and the forehead regions are therefore encoded using the TSVQ technique whereas the eyes regions are encoded using the BTC technique. It, therefore, takes advantage of the merits of both the BTC and the TSVQ methods to achieve a high quality and high compression ratios of the images. From experimental results, the proposed combined method is found to yield better image quality than the TSVQ technique at the same codeword length and block size. Its compression ratio is also higher than that of the BTC technique at the same block size.