Accelerating Image Processing Algorithms Based on the Reuse of Spatial Patterns

This paper presents window memoization, a performance optimization technique for convolution-based image processing algorithms. Window memoization exploits the repetitive nature of image data to reduce the number of calculations required for image processing algorithms and hence, it improves the performance. We applied window memoization to a chain of image processing algorithms that includes median filter, Kirsch edge detector and local edge filling. We found that a large portion of the calculations performed on pixel neighborhoods can be skipped and instead, previously calculated results can be reused. The typical speedups were in the range of 1.6times to 2.8times.