Low power DCT implementation using differential pixels for on-board satellite image processing

Abstract

Low power and high speed discrete cosine transform (DCT) implementation of the images captured by the satellites presents a hardware design problem. The cost of the DCT implementation is dominated by the complexity of the multiplication of input data (image) with the DCT matrix. This paper presents the techniques for minimizing the complexity of multiplication by employing differential pixel image (DPI). In our method we take 8times8 blocks of input image matrix and calculate the difference between the adjacent pixels and use those differential pixels in DCT transformation. While conventional low complexity DCT implementation methods focus on optimizing the DCT matrix, our method achieves complexity reduction by using the difference of intensities of adjacent pixels of the image. Design examples show that an average of 12% reduction in power consumption and delay is achieved over the conventional method. Synthesis results on 0.18 mum CMOS technology show that our method gives an average of 13.2% reduction in power consumption and 10.9% improvement in speed over the conventional method.