Cache memory loaclity optimization for implementation of computer vision and image processing algorithms

Abstract EN

One of the main problems in developing fast image processing and computer vision systems is the memory speed.

Memory speed represents the performance bottleneck due to the large gap between processor and memory speeds.

Cache memory is very fast, but it is small to store all required data and instructions.

In this paper, image processing and computer vision algorithms are optimized to enhance performance by increasing the cache memory utilization.

This optimization increases the spatial locality and temporal locality and improves the system performance.

The proposed optimization is applied on a set of image processing operations such as image intensity transformation, image filtering, geometric transformation, and CNN.

The time analysis of the systems has shown a speed improvement of 30% to 70% compared with direct algorithm implementation.