Traditionally, the edge detection process requires one final step that is known as scaling. This is done to decide, pixel by pixel, if these will be selected as final edge or not. This can be considered as a local evaluation method that presents practical problems, since the edge candidate pixels should not be considered as independent. In this article, we propose a strategy to solve these problems through connecting pixels that form arcs, that we have called segments. To accomplish this, our edge detection algorithm is based on a more global evaluation inspired by actual human vision. Our paper further develops ideas 1 first proposed in Venkatesh and Rosin (Graph Models Image Process 57(2):146-160, 1995). These segments contain visual features similar to those used by humans, which lead to better comparative results against humans. In order to select the relevant segments to be retained, we use fuzzy clustering techniques. Finally, this paper shows that this fuzzy clustering of segments presents a higher performance compared to other standard edge detection algorithms.
KeywordsEdge detection • Global evaluation • Supervised classification • Fuzzy clustering • Edge segments Edge detection is quite useful in many fields. For instance, for the recognition of different pathologies in medical diagnoses (Sonka 2014), a field has grown in recent years. It is also used in images taken by satellites or drones-remote sensing-for agricultural purposes. Some other relevant fields of application are the military industry, law enforcement, among others (Monga et al. 1991; Fathy and Siyal 1995; Zielke et al. Communicated by I. Perfilieva.
The majority of edge detection algorithms only deal with grayscale images, while their use with color images remains an open problem. This paper explores different approaches to aggregate color information of RGB and HSV images for edge extraction purposes through the usage of the Sobel operator and Canny algorithm. This paper makes use of Berkeley’s image data set, and to evaluate the performance of the different aggregations, the F-measure is computed. Higher potential of aggregations with HSV channels than with RGB channels is found. This article also shows that depending on the type of image used, RGB or HSV, some methods are more appropriate than others.
Traditionally, the last step of edge detection algorithms, which is called scaling-evaluation, produces the final output classifying each pixel as edge or nonedge. This last step is usually done based on local evaluation methods. The local evaluation makes this classification based on measures obtained for every pixel. By contrast, in this work, we propose a global evaluation approach based on the idea of edge list to produce a solution that suits more with the human perception. In particular, we propose a new evaluation method that can be combined with any classical edge detection algorithm in an easy way to produce a novel edge detection algorithm. The new global evaluation method is divided in four steps: in first place we build the edge lists, that we have called edge segments. In second place we extract the characteristics associated to each segment: length, intensity, location, and so on. In the third step we learn the characteristics that make a segment good enough to become an edge. At the fourth step, we apply the classification task. In this work we have built the ground truth of edge list necessary for the supervised classification. Finally, we test the effectiveness of this algorithm against other classical algorithms based on local evaluation approach.
In this paper, machine learning (ML) techniques are applied at an early stage of Image Processing (IP). The learning procedures are usually applied from at least the image segmentation level, whereas, in this paper, this is done from a lower processing level: the edge detection level (ED). The main objective is to solve the edge detection problem through ML techniques. The proposed methodology is based on a classification of edges made pixel by pixel, but the predictors employed for the ML task include information about the pixel neighborhood and structures of connected pixels called edge segments. The Sobel operator is employed as input. Making use of 50 images that belong to the Berkeley Computer Vision data set, the average performance of the validation sets when employing our Neural Networks method reached an F-measure significatively higher than with the Sobel operator. The experiment results show that our post-processing technique is a promising new approach for ED.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.