Edge Detection with the Cross-layer Fusion Feature Based on RCF

To solve the problems such as chaotic and fuzzy edge lines caused by current edge detection based on deep-learning, inspired by RCF (Richer Convolutional Features), an end-to-end Cross-layer Fusion Feature for edge detection (CFF) model was presented. In it, RCF was used as a baseline, CBAM (Convolutional Block Attention Module) was appended to the backbone network, translation-invariant downsampling technology was adopted, and some downsampling operations in the backbone network were deleted to preserve the image details information, dilated convolution technique was adopted to increase the model receptive field simultaneously. Additionally, the cross-layer fusion feature map is used to fully fuse high level and low-level features together. To balance the relationship between each stage loss and the fusion loss, and to avoid the phenomenon of excessive loss of low-level details after multi-scale feature fusion, the weight parameters were added to the losses. The model was trained on the PASCAL VOL Context and Berkeley Segmentation (BSDS500) Set, and the image pyramid technology was used in testing to improve the edge images quality. Experimental results verified that the contour extracted by CFF model is clearer than that extracted by the baseline network and can improv the edge blurring problem.