Optimization methods for edge extraction in digital images

Abstract

Edge detection plays a crucial role in identifying boundaries and key features in images, providing valuable information for various applications such as object recognition, image segmentation, pattern analysis, and stereoscopic vision. However, this process, often required in real-time, involves a high computational demand, which has driven significant efforts toward optimization. This article compares two optimization methods to efficiently address this task: the optimization method using SIMD instructions and the metaheuristic Artificial Bee Colony method modified for edge detection (ABC-ED). First, the SIMD instruction method is implemented, performing parallel edge inspection by simultaneously processing four pixels and optimizing resource usage. Next, the ABC-ED method is developed, highlighting its effectiveness in detecting edges by evaluating less than one-third of the image’s pixels. Nevertheless, despite narrowing the search region, it still exhibits a high execution time. In contrast, the SIMD instruction method demonstrates efficiency and speed in edge detection, positioning it as the most suitable option for this task. A comparison with other state-of-the-art methods reveals superior performance of the SIMD instruction method, even when compared to implementations in the well-known OpenCV open-source library aimed at the same task. Code is available at https://github.com/elimm1910/OptimizationMethodsEdgeDetect.