Dense Extreme Inception Network: Towards a Robust CNN Model for Edge Detection

Abstract

Edge detection is the basis of many computer vision applications. State of the art predominantly relies on deep learning with two decisive factors: dataset content and network's architecture. Most of the publicly available datasets are not curated for edge detection tasks. Here, we offer a solution to this constraint. First, we argue that edges, contours and boundaries, despite their overlaps, are three distinct visual features requiring separate benchmark datasets. To this end, we present a new dataset of edges. Second, we propose a novel architecture, termed Dense Extreme Inception Network for Edge Detection (DexiNed), that can be trained from scratch without any pretrained weights. DexiNed outperforms other algorithms in the presented dataset. It also generalizes well to other datasets without any fine-tuning. The higher quality of DexiNed is also perceptually evident thanks to the sharper and finer edges it outputs.

Colour metrics for image edge detection

“…Three datasets including BSDS500, MBDD, and BIPED were used in the experimental section for quantitative evaluation of new method. BSDS500 contains 500 images of which 300 of images are used for training and validation, the remaindings 200 images are used for testing.…”

“…BSDS500 contains 500 images of which 300 of images are used for training and validation, the remaindings 200 images are used for testing. Every image in BSDS is annotated at least by six annotators; this dataset is mainly intended for image segmentation and boundary/contour detection . The boundary/contour detection tasks, although related to edge detection but are different.…”

“…BIPED is a carefully annotated edge dataset. It contains 250 outdoor images of 1280 × 720 pixels, 50 of the images are used for testing …”

“…Three evaluation metrics which usually used as quantitative metrics are used: ODS which employs a fixed threshold for all images in a dataset, OIS which selects an optimal threshold for each image, and average precision (AP). The F ‐measure ( F ) of ODS and OIS, will be considered, where F is calculated using Equation …”

“…Edge detectors are a collection of important local image processing methods designed to detect edge pixels . Edge detection methods could be categorized as: (a) low level feature; (b) brain‐biologically inspiration; (c) classical learning algorithms; (d) deep‐learning algorithms . Some “low level feature” based edge detectors for gray scale images understanding have been developed over the past few decades.…”