International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056 Volume: 04 Issue: 01 | Jan -2017 www.irjet.net p-ISSN: 2395-0072 © 2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 1391 An Analysis of Energy Efficient Gaussian Filter Architectures Mrinal Dubey 1 and Shweta Agrawal 2 1 Research Scholar, 2 Assistant Professor, Dept. of Electronics and Comm., SRCEM Banmore, Morena, India ---------------------------------------------------------------------***--------------------------------------------------------------------- Abstract - The high usage of the portable multimedia devices such as camera, phones etc. demanding highly compute intensive operations. Therefore, energy efficient design techniques are prime requirement to reduce the battery lifetime. To achieve energy efficient designs several approaches have been proposed. Among the various techniques, the approximate computing has emerged as the promising alternative to the conventional approaches and provides very good energy efficiency. The Gaussian filter is most commonly used the pre-processing filter before edge detection to remove the unwanted edges due to noise. Most of the existing architectures for energy efficient Gaussian filtering include approximation of kernel coefficients, change of window size, architectural modification and employing approximate arithmetic in the accurate design. This paper reviews different energy efficient filter architectures and compares them. Key Words: Image Processing, Gaussian Filter, Integrated Circuits, Quality Tunable designs. 1. INTRODUCTION Recently, there is increasingly usage of the portable devices employing multimedia applications such as camera, phones etc. demanding highly compute intensive applications on battery operated devices. Therefore, energy efficient design techniques are becoming more popular to reduce the battery lifetime [1]. In order to achieve energy efficient designs several approaches from algorithm/architecture to circuit level have been proposed. Among the various techniques, the approximate computing has emerged as the promising alternative to the conventional approaches and provides very good energy efficiency for the error tolerant applications. Since most image processing applications produces output for human consumption which exhibits limited perception, small amount of error in the output image cannot be discerned. All these applications are commonly called as error tolerant applications. In these applications, small amount of error is acceptable. Therefore, approximate designs are developed for various image processing computing cores. The edge detection is the frequently used operation in several computer vision applications. Since the Gaussian filtering (GF) the most compute intensive operation within edge detection, energy efficient design of GF is prime requirement. The GF is the weighted non-linear filter and is used to blur and de-noise the image. It is most commonly used the pre-processing filter before edge detection to remove the unwanted edges due to noise [2], [3]. To efficiently implement the GF, the Gaussian equation is approximated by a kernel of different sizes. Larger the size of kernel better is the approximation of Gaussian expression that provides higher quality at the cost large computational complexity. In order to achieve filtered image, convolution of the Gaussian kernel with the image sub-matrix is performed. Since the accurate kernel exhibits floating point constant multiplier, it consumes large energy. Most of the existing GF filter design exhibits approximation of kernel coefficients [4], change of window size, architectural modification and employing approximate arithmetic in the accurate design. The floating point kernel coefficients are approximated to fixed point and sum of power of two form to reduce the implementation complexity [5]. Further, different window size of kernel is considered for obtaining quality energy trade-off. Furthermore, some approaches modify the architectures for obtaining energy scalable Gaussian filtering [6], [7]. Finally, some of the energy efficient architectures are obtained by embedding approximate adder [8], [9]. These architectures provide quality energy trade-off. The remainder of this paper explores different Gaussian filters with comparative analysis by implementation and simulation.