Image coding using vector quantization based on wavelet transform fuzzy C-Means and principle component analysis

  • Authors:

  • Tanasak Phanprasit;Thurdsak Leauhatong;Chuchart Pintavirooj;Manas Sangworasil

  • Affiliations:

  • Department of Electronics and Telecommunication Engineering, Bangkok University, Patumthani, Bangkok, Thailand and Faculty of Engineering, King Mongkut's Institute of Technology Ladkrabang, Bangko ...;Faculty of Engineering, King Mongkut's Institute of Technology Ladkrabang, Bangkok, Thailand;Faculty of Engineering, King Mongkut's Institute of Technology Ladkrabang, Bangkok, Thailand;Faculty of Engineering, King Mongkut's Institute of Technology Ladkrabang, Bangkok, Thailand

  • Venue:
  • ISCIT'09 Proceedings of the 9th international conference on Communications and information technologies
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

Image coding requires a small bit rate for high-speed data transmission and a small space for data storage. Simultaneously, the peak signal to noise ratio (PSNR) has to be maintained. In this paper, we proposed a method of image coding design using wavelet transform (WT). By applying the WT for defining groups of pixels with the same intensity in spatial domain, the groups of pixels are allocated in a low frequency range. Hence, locations of pixels are the key factor to determine the size of each block and we use wavelet transform to decompose each block into subband components, which are represented by 3D vectors. The 3D vectors are then classified into 8 groups corresponding to quadrants of spatial coordinates. In addition, we apply Fuzzy C-Means algorithm to classify the member in the magnitudes value of 3D vectors into code vector. Due to the lossy coding process, we propose a method of system error compensation on Vector Quantization (VQ) by using principle component analysis and discrete wavelet transform to performed on the system error and keeping the high-energy coefficient for further inverse wavelet transform to yield system error compensation. The reconstructed image and system error compensate will be combined in order to construct an output image (Xo). By applying the proposed method, performance of the method is evaluated as 26.19% of bit rate and 1.50% of PSNR improved.