An image segmentation method for steel reinforcement recognition

Abstract:
In this paper, by improving the classical Otsu threshold segmentation algorithm of threshold selection function, the infra-class variance is taken into the consideration of the best threshold value calculation to get closer to the ideal threshold. It solves the problem that the traditional Otsu threshold segmentation method can not process the image well when the contrast is low. Especially when processing images in places with high noise and low contrast, it can segment images well. For the end image of reinforcing steel, the smaller the infra-class variance is, the greater the inter-class variance is, and the resulting threshold is more close to the ideal threshold, the better is the effect of Image segmentation is. After using OpenCV to process the image, it can be clearly seen that the average error detection rate of the segmentation result of the improved method is 0.026, while the average error detection rate of the segmentation result of the traditional method is 0.251. The improved algorithm is much better than the original algorithm, nearly 10 times. Therefore, this paper puts forward an improved Otsu threshold selection function, which can take into account of the impact of both of the infraclass variance and the inter-class variance.