Approximate Newton method using frequency upscaling for total variation-based image denoising

Approximate computing is an effective low-power technique that improves circuits’ energy and latency performance by allowing acceptable output errors in the image processing design. This paper presents a runtime-based approximate integrated method by applying frequency upscaling (FUS) into the approximate Newton method for total variation (TV)-based image denoising. The proposed technique increases the processing rate of the approximate Newton method by increasing the frequency of input vectors beyond the maximum correct frequency for adder cells; the number of output errors varies without changing the circuit structure. The results illustrate that the number of completions for the algorithm increases when the frequency increases gradually. The input frequency (22.64GHZ) that the approximate full adder sustains is higher than that for exact full adder (21.04GHZ) by maintaining a low number of iterations in the approximate Newton method. The output image quality remains nearly unchanged for both exact and approximate cells when the number of approximate bits (NAB) is smaller than 18.