The Algorithm and VLSI Architecture of an Efficient Image Sharpening Scheme Based on the Frequency Domain Analysis

Abstract

Feature augmentation of the images plays a big role in the image recognition system and using a sharper image can enhance the accuracy of the image recognition. This paper presents a non-iterative image sharpening algorithm based on the frequency-domain analysis. This algorithm greatly improves the noise sensitive issue in classical unsharp masking technique. Furthermore, the non-iterative property is conducive to the hardware acceleration design. Considering the design of hardware acceleration, we focus on the three parts with the highest computation load of the algorithm, including FFT/IFFT unit, Wiener deconvolution process, and the least square method. Through the hardware acceleration, this algorithm is more suitable for applying to the real-time image recognition system. The experiment results show that this algorithm enhances the feature and contrast of image with effective noise reduction and achieves a throughput 30fps.