Vector quantisation using hybrid teaching learning and pattern search optimisation for image compression

Abstract

The advantage of vector quantisation (VQ) using hybrid teaching learning-based optimisation and pattern search (hTLBO-PS) for image compression over ant colony and firefly techniques is validated. Further, the efficiency of proposed hTLBO-PS on PSNR and reconstructed image quality over other optimisation is verified. The first efficient VQ is Linde-Buzo-Gray (LBG), generates a local optimal codebook, but undergoes local optimal problem and lower PSNR values. So researchers proposed ant colony and firefly algorithm for global optimisation, but undergoes problems because of considerable tuning parameters and no such brighter firefly in the search respectively. From the literature, TLBO is good in exploitation and pattern search is good in exploration, so we hybridise the TLBO and PS based on their strengths and weakness. Finally, the performance hTLBO-PS is compared with the ACO and FA and proved better in efficient codebook design leads to higher peak signal to noise ratio and excellent reconstructed image quality. The optimised vector quantised codebook and index table are coded through run-length coding followed by Huffman coding at the transmitter section, whereas at receiver section the original image is reconstructed by decoders of Huffman and run-length.