A Novel Hybrid Deep Learning Model for Image Edge Detection

Image processing is an emerging field in this digital world as digital content holds a significant number of interpretations at a high level as well as ground level. Edge detection plays a vital role in many applications such as healthcare, business, inventory and so on. As a result, the number of techniques to identify the edges in images is high. However, the significance of the present techniques is at the debate level. Hence in this approach, two conventional image edge detection methods namely Sobel and Robert are hybridized with CNN architecture to improve the performance of edge detection. After the deep learning approach, the fuzzy technique was used to enhance the identified edges in the image. The experimentation part shows the significance of the proposed approach and the proposed SR-CNN is compared with four other existing methods. For experimentation, 6 images are used for evaluating the proposed SR-CNN. On comparing the results of the proposed method SR-CNN, it surpasses the Canny with 68.29%, Prewitt with 82.19%, Sobel with 83.33% and Laplacian with 83.12%, On an average, the entire set of methods are surpassed by proposed SR-CNN by 79.23%.