An Adaptive Threshold Setting Algorithm Based on PCNN Edge Detection Model
Xiangyu Deng, Yahang Yang, Haijun Sun · 2021
Edge detection is an important image preprocessing technology, which plays an important role in image classification, target recognition and image understanding, but the parameter setting of various edge detection algorithms is always a difficult problem. This paper aims at the problem that the segmentation threshold in the PCNN edge detection model proposed in the previous stage needs to be manually set, utilizing the characteristics of the image edges contained in the PCNN segmentation output, through the combination of transform domain edge feature extraction and DNN prediction, realize the adaptive setting of the segmentation threshold parameters of the PCNN detection model. First, use Contourlet transform to extract sub-band feature vectors on the output of each iteration of the PCNN segmentation model, further perform median pooling processing on the extracted feature vectors, and then calculate the variance and mean value as the feature vector, and then propose an edge detection algorithm that uses DNN to realize segmentation threshold adaptive prediction. When predict the appropriate segmentation threshold, this method can get better edge detection effect of the image only after six iteration of PCNN. When the algorithm is applied to images such as Lena and BSDS500 image library, it has obtained a finer edge detection effect and has high robustness.