<![CDATA[Edge detection is a critical stage in digital image processing, where the quality of object boundary representation directly determines the accuracy of subsequent processes such as segmentation and pattern recognition. This study compares three edge detection operators—Sobel, Prewitt, and Canny—on five digital images of fruit objects based on Python thru Google Colab, with an analytical emphasis on the influence of Gaussian Blur preprocessing and variations in Canny threshold parameters on the quality of detection results. Unlike previous comparative studies, this research specifically analyzes the methodological implications of using MSE, PSNR, and SSIM in edge image evaluation—whose distribution characteristics fundamentally differ from natural images—and provides guidance on selecting Canny thresholds based on systematic testing on complex organic contour images. The evaluation was conducted thru a combination of visual analysis and quantitative measurements of Mean Squared Error (MSE), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity Index (SSIM). The results show that the Canny operator produces the best visual edge detection quality across all test images, despite having a higher average MSE value. This paradox is explained by the sparse output nature of Canny: most pixels have a value of zero, causing the deviation from the reference grayscale image to increase in aggregate, not due to inferior quality. Threshold configuration (100, 200) with a T_{high} /T_{low}\ ratio of 2 provides an optimal balance. These findings affirm that the evaluation of edge detection methods requires a multiparameter approach and careful contextual interpretation, rather than relying solely on pixel-level metrics.]]>
Copyrights © 2026
