Morphological Edge Detection Algorithms on the Noisy Car Image Database

Abstract

Development of edge detector using mathematical morphology can provide remarkably more precise edge detection. This system mainly focuses on edge detection of cars. This paper uses car database for the early detection in automatic traffic law enforcement. The methodology for accurate edge detector includes grayscale image conversion, median filter, element structures formation, mathematical morphology, synthetic weighted and segmentation using Otsu method. The morphological operation that was used in this study is basic operation such as dilation and erosion. The use of morphological operation was to find the limits and the image terminals by selecting different size and extensions converted to images of grey levels and the results were compared to determine the most suitable and better methods. In this study, 371 car databases and 100 objects from the publicly available Berkeley and Greek car databases were used as data. This methodology is proved highly accurate 0.7-1.58% higher than Canny edge detector. The accuracy of the results from Mathematical Morphlogy Edge Detection were 83.23%; 83.27%; 82.66%; 79.78% while the results from Canny Edge Detection were 81.90%; 82.08%; 81.08%; 79.71%. In comparison with the results from Canny edge detector. It shows that mathematical morphology renders better overall performance. It was also tested with salt and pepper noises and still shows better results. Z-test was used for comparing the means of two populations while F-test was used to test if two population variances are equal. Both tests were done because in this study two populations were used as main datasets. The effectiveness and robustness make this mathematical morphology method a suitable tool to be integrated into complete pre-screening systems for the early detection in automatic traffic law enforcement.