An Approximate Discrete Hadamard Transform for Energy Efficient Multimedia Processing

Many of the applications that are used in the handheld devices are multimedia processing. Multimedia processing will consume huge power due to its computations, as a result battery gets drained quickly. Multimedia applications are error tolerant applications, as human eye can not able to recognize the error present in the multimedia content. Approximate computing is one of the alternative to enhance the battery life. Hadamard transform is used in image coding, CDMA and MPEG, etc,. applications. This paper proposes an approximate hardware architecture for computing discrete Hadamard transform. The proposed approximate Hadamard transform architecture consumes 33% and 31 % lesser area, 22% and 14% lesser delay, power reduces by 36% and 39% compared to the existing exact 16-point and 4-point Hadamard transform. The PSNR values of 4-bit LSB approximate 16-point and 4-point Hadamard transform's output image is 36.30 and 36.32 respectively.