FCL-Net: Towards accurate edge detection via Fine-scale Corrective Learning

Xuan, Wenjie; Huang, Shaoli; Liu, J.; Du, Bo

doi:10.1016/j.neunet.2021.10.022

Cited by 31 publications

(4 citation statements)

References 27 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In order to address the aforementioned drawbacks, Deng et al ( [7,6]) used the decoder structure of U-Net [34] to gradually incorporate global information into the shallow features. However, recent research [49] suggests that semantic information gradually decays as it is fused down-ward in U-Net structures, diminishing its guiding effect. To simultaneously preserve multi-scale features and generate pixel-level weight matrices, recent works such as FCL [49] generate a pixel-level weight matrix for each scale feature during multi-scale feature generation.…”

Section: Edge Fusion Methodsmentioning

confidence: 99%

“…However, recent research [49] suggests that semantic information gradually decays as it is fused down-ward in U-Net structures, diminishing its guiding effect. To simultaneously preserve multi-scale features and generate pixel-level weight matrices, recent works such as FCL [49] generate a pixel-level weight matrix for each scale feature during multi-scale feature generation. CATS [17], on the other hand, splices multi-scale features and combines spatial and channel information to alleviates edge localization ambiguity.…”

Section: Edge Fusion Methodsmentioning

confidence: 99%

See 1 more Smart Citation

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

Soria

Riba

Sappa

2020

2020 IEEE Winter Conference on Applications of Computer Vision (WACV)

184

122

View full text Add to dashboard Cite

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.

show abstract

Section: Edge Fusion Methodsmentioning

confidence: 99%

Section: Edge Fusion Methodsmentioning

confidence: 99%

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

Soria

Riba

Sappa

2020

2020 IEEE Winter Conference on Applications of Computer Vision (WACV)

184

122

View full text Add to dashboard Cite

show abstract

“…A decoder structure of U-Net [ 22 ] was used in [ 23 ] to incorporate global information into shallow features. However, a recent study [ 24 ] suggested that semantic information gradually decays as it is fused downward in U-Net structures. An edge detection approach [ 25 ] based on U-Net performed poorly on the BSDS500 dataset.…”

Section: Related Workmentioning

confidence: 99%

Enhancing Deep Edge Detection through Normalized Hadamard-Product Fusion

Hu,

Saeli

2024

J. Imaging

View full text Add to dashboard Cite

Deep edge detection is challenging, especially with the existing methods, like HED (holistic edge detection). These methods combine multiple feature side outputs (SOs) to create the final edge map, but they neglect diverse edge importance within one output. This creates a problem: to include desired edges, unwanted noise must also be accepted. As a result, the output often has increased noise or thick edges, ignoring important boundaries. To address this, we propose a new approach called the normalized Hadamard-product (NHP) operation-based deep network for edge detection. By multiplying the side outputs from the backbone network, the Hadamard-product operation encourages agreement among features across different scales while suppressing disagreed weak signals. This method produces additional Mutually Agreed Salient Edge (MASE) maps to enrich the hierarchical level of side outputs without adding complexity. Our experiments demonstrate that the NHP operation significantly improves performance, e.g., an ODS score reaching 0.818 on BSDS500, outperforming human performance (0.803), achieving state-of-the-art results in deep edge detection.

show abstract

“…For instance, in image segmentation for boundary detection, coastlines extraction between sea and land regions in aerial images, detection of buildings, content-based image retrieval, and other machine vision routines (Stoian et al, 2019). (Xuan et al, 2022). Consequently, numerous algorithms that work with edge recognition can identify various forms of edges and these algorithms can be considered as a distinct method depending on the for instance, it may be a limit that separates an image region into discontinuity, or outlines that can form the object boundaries.…”

Section: Introductionmentioning

confidence: 99%

Multi-Phase Information Theory-Based Algorithm for Edge Detection of Aerial Images

Elaraby

Al-Ameen

2022

Journal of Information and Communication Technology

View full text Add to dashboard Cite

Edge detection is the diverse way used to detect boundaries in digital images. Many methods exist to achieve this purpose, yet not all of them can produce results with high detection ratios. Some may have high complexity, and others may require numerous inputs. Therefore, a new multi-phase algorithm that depends on information theory is introduced in this article to detect the edges of aerial images adequately in a fully automatic manner. The proposed algorithm operated by utilizing Shannon and Hill entropies with specific rules along with a non-complex edge detector to record the vital edge information. The proposed algorithm was examined with different aerial images, its performances appraised against six existing approaches, and the outcomes were assessed using three image evaluation methods. From the results, promising performances were recorded as the proposed algorithm performed the best in many aspects and provided satisfactory results. The results of the proposed algorithm had high edge detection ratios as it was able to capture most of the significant edges of the given images. Such findings make the proposed algorithm desirable to be used as a key image detection method with other image-related applications.

show abstract

FCL-Net: Towards accurate edge detection via Fine-scale Corrective Learning

Cited by 31 publications

References 27 publications

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

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

Enhancing Deep Edge Detection through Normalized Hadamard-Product Fusion

Multi-Phase Information Theory-Based Algorithm for Edge Detection of Aerial Images

Contact Info

Product

Resources

About