Real-time video thresholding using evolutionary techniques and cross entropy

Evolutionary Algorithms (EAs) are present in most areas of science and engineering where difficult problems arise. However, EAs are often applied to design problems where the speed is not a crucial factor. This tendency has lead EAs to be excluded from real-time applications due to its iterative nature. Image processing has benefited from EAs on many off-line applications, but little research has been made for real-time image processing problems. This paper presents the evaluation of EAs applied to the thresholding of a stream of images in real-time. Results indicate that Differential Evolution (DE) can be modified to achieve real-time performance on a single core implementation without any form of parallelization. These circumstances indicate that the performance can be further improved with multi-core implementations or GPU parallelization.