Berkeley Segmentation Benchmark Research Articles

Multi-level Image Segmentation using Randomized Spiral-based Whale Optimization Algorithm

Background: Segmenting an image into multiple regions is a pre-processing phase of computer vision. For the same, determining an optimal set of thresholds is a challenging problem. Objective: This paper introduces a novel multi-level thresholding based image segmentation method. Methods: The presented method uses a novel variant of whale optimization algorithm to determine the optimal thresholds. For experimental validation of the proposed variant, twenty-three benchmark functions are considered. To analyze the efficacy of new multi-level image segmentation method, images from Berkeley Segmentation Dataset and Benchmark (BSDS300) have been considered and tested against recent multi-level image segmentation methods. Conclusion: Experiments arm that the presented method is efficient and competitive than the existing multi-level image segmentation methods.

A multi-feature bipartite graph ensemble for image segmentation

Image segmentation benefits from using multi-feature ensembles. In this paper, we propose a novel multi-layer bipartite graph model for more effective feature fusion. This model employs multiple graph layers, each representing a feature space. They share common vertices but have individual edge sets that are obtained from different feature spaces. The features are fused by the regularization defined on a Grassmann manifold, which compresses the graph layers into subspace representations and merges them into one. The merged graph is then fed into a graph-cut algorithm to generate the final segmentation. Experiments carried out using the Berkeley segmentation benchmark show that our model is effective in feature fusion and image segmentation, outperforming the state-of-the-art superpixel-based methods.

Superpixels extracted via region fusion with boundary constraint

In this paper, we present an accurate superpixel algorithm by region fusion with boundary constraint (RFBC). Superpixels with regular shape and high boundary adherence can be generated in weak boundary and complex texture regions through our algorithm. RFBC includes two steps which are initial segmentation and region fusion respectively. In initial segmentation, broken Canny edges are connected through edge closing algorithm. Subsequently, the closed Canny edges and SLIC superpixel edges are combined together to form the incipient superpixels. In region fusion, gray Gaussian distribution and adjacent relation are used as priori to compute the degree of similarity across incipient superpixels in GBP algorithm. For concreteness, the information of similarity is propagated between regions and the most similar regions are fused, which are accomplished alternatingly to preserve accurate boundaries. Extensive experiments on the Berkeley segmentation benchmark show that the proposed algorithm outperforms the most state-of-the-art algorithms.

Contour detection based on anisotropic edge strength and hierarchical superpixel contrast

Contour detection is a fundamental problem in computer vision, yet existing methods usually suffer from the interference of noise and textures. To address this problem, we present an unsupervised contour detection method based on anisotropic edge strength and hierarchical superpixel contrast. The anisotropic edge strength is obtained through the first derivative of anisotropic Gaussian kernels which incorporates an adaptive anisotropy factor. The anisotropic kernel improves the robustness to noise, while the adaptive anisotropic factor attenuates the anisotropy stretch effect. Using a method based on region merging, we obtain a hierarchical set of superpixel maps and thus compute superpixel contrast maps at different hierarchy levels. Consequently, the contour strength map is obtained by multiplying the anisotropic edge strength map by the average of the hierarchical superpixel contrast maps. Experimental results on two publicly available datasets validate the superiority of the proposed method over the competing methods. On the Berkeley Segmentation Dataset & Benchmark 300 and the Berkeley Segmentation Dataset & Benchmark 500, our method obtains (optimal dataset scale) F-measure values of 0.63 and 0.67, respectively, an improvement of at least 0.06 over the competing methods.

FLIC: Fast linear iterative clustering with active search

In this paper, we reconsider the clustering problem for image over-segmentation from a new perspective. We propose a novel search algorithm called “active search” which explicitly considers neighbor continuity. Based on this search method, we design a back-and-forth traversal strategy and a joint assignment and update step to speed up the algorithm. Compared to earlier methods, such as simple linear iterative clustering (SLIC) and its variants, which use fixed search regions and perform the assignment and the update steps separately, our novel scheme reduces the number of iterations required for convergence, and also provides better boundaries in the over-segmentation results. Extensive evaluation using the Berkeley segmentation benchmark verifies that our method outperforms competing methods under various evaluation metrics. In particular, our method is fastest, achieving approximately 30 fps for a 481 × 321 image on a single CPU core. To facilitate further research, our code is made publicly available.

FLIC: Fast Linear Iterative Clustering With Active Search

In this paper, we reconsider the clustering problem for image over-segmentation from a new perspective. We propose a novel search algorithm named “active search” which explicitly considers neighboring continuity. Based on this search method, we design a back-and-forth traversal strategy and a "joint" assignment and update step to speed up the algorithm. Compared to earlier works, such as Simple Linear Iterative Clustering (SLIC) and its follow-ups, who use fixed search regions and perform the assignment and the update step separately, our novel scheme reduces the iteration number before convergence, as well as improves boundary sensitivity of the over-segmentation results. Extensive evaluations on the Berkeley segmentation benchmark verify that our method outperforms competing methods under various evaluation metrics. In particular, lowest time cost is reported among existing methods (approximately 30 fps for a 481321 image on a single CPU core). To facilitate the development of over-segmentation, the code will be publicly available.

Learning and surface boundary feedbacks for colour natural scene perception

Boundary detection and segmentation are essential stages in object recognition and scene understanding. In this paper, we present a bio-inspired neural model of the ventral pathway for colour contour and surface perception, called LPREEN (Learning and Perceptual boundaRy rEcurrent dEtection Neural architecture). LPREEN models colour opponent processes and feedback interactions between cortical areas V1, V2, V4, and IT, which produce top-down and bottom-up information fusion. We suggest three feedback interactions that enhance and complete boundaries. Our proposed neural model contains a contour learning feedback that enhances the most probable contour positions in V1 according to a previous experience, and generates a surface perception in V4 through diffusion processes. We compared the proposed model with another bio-inspired model and two well-known contour extraction methods, using the Berkeley Segmentation Benchmark. LPREEN showed better performance than two methods and slightly worse performance than another one.

Fast Edge Detection Using Structured Forests

1 B-404,Neelsidhi Atlantis, PIIT, Navi Mumbai, India 2 Sector 19A,Nerul, New Panvel, Navi Mumbai, India Email: bindu_k_nair@hotmail.com, ujwal.harode@gmail.com ABSTRACT An important and very crucial component of many systems involving images is the edge detection which includes object detectors and image segmentation algorithms. Edge patches always shows shapes of its inherent structure, like T junctions or lines. Taking advantage of the shapes available in the images we get to know of an edge detector which is accurate and efficient. In this approach the structured labels is robustly mapped to a space that is discrete by which evaluation of the measure of the standard information gain is possible. The outcome is a way that shows a real time performance which is many times faster than any of the traditional and modern approaches available today, which also achieves the best results for edge detection both the Berkeley Segmentation Dataset and Benchmark (BSDS 500) and the NYU depth dataset. This approach also shows its strength as a an edge detector that can be used for any regular purpose by displaying how this method generalize across all datasets

Hierarchical Image Segmentation Using a Combined Geometrical and Feature Based Approach

This paper presents a fully automatic segmentation algorithm based on geometrical and local attributes of color images. This method incorporates a hierarchical assessment scheme into any general segmentation algorithm for which the segmentation sensitivity can be changed through parameters. The parameters are varied to create different segmentation levels in the hierarchy. The algorithm examines the consistency of segments based on local features and their relationships with each other, and selects segments at different levels to generate a final segmentation. This adaptive parameter variation scheme provides an automatic way to set segmentation sensitivity parameters locally according to each region's characteristics instead of the entire image. The algorithm does not require any training dataset. The geometrical attributes can be defined by a shape prior for specific applications, i.e. targeting objects of interest, or by one or more general constraint(s) such as boundaries between regions for non-specific applications. Using mean shift as the general segmentation algorithm, we show that our hierarchical approach generates segments that satisfy geometrical properties while conforming with local properties. In the case of using a shape prior, the algorithm can cope with partial occlusions. Evaluation is carried out on the Berkeley Segmentation Dataset and Benchmark (BSDS300) (general natural images) and on geo-spatial images (with specific shapes of interest). The F-measure for our proposed algorithm, i.e. the harmonic mean between precision and recall rates, is 64.2% on BSDS300, outperforming the same segmentation algorithm in its standard non-hierarchical variant.

Extraction of flower regions in color images using ant colony optimization

Abstract Extraction of flower regions from complex background is a difficult task and it is an important part of a flower image retrieval and recognition. In this article, we propose an Ant Colony Optimization (ACO) algorithm as a general color clustering method, and test it on flower images as a case study of object boundary extraction. The segmentation methodology on flower images consists of six steps: color space conversion, generation of candidate color cluster centers, ant colony optimization method to select optimum color cluster centers, merging of cluster centers which are close to each other, image segmentation by clustering, and extraction of flower region from the image. To evince that ACO algorithm can be a general segmentation method, some results of natural images in Berkeley segmentation benchmark have been presented. The method as a case study on flower region extraction has also been tested on the images of Oxford-17 Flowers dataset, and the results have confronted with other well established flower region extraction approaches.

Edge detection in the feature space

To build a consistent image representation model which can process the non-Gaussian distribution data, a novel edge detection method (KPCA-SCF) based on the kernel method is proposed. KPCA-SCF combines kernel principal component analysis and kernel subspace classification proposed in this paper to extract edge features. KPCA-SCF was tested and compared with linear PCA, nonlinear PCA and conventional methods such as Sobel, LOG, Canny, etc. Experiments on synthetic and real-world images show that KPCA-SCF is more robust under noisy conditions. KPCA-SCF's score of F-measure (0.44) ranks 11th in the Berkeley segmentation dataset and benchmark, it (0.54) ranks 10th tested on a noised image.

Texture Analysis and Segmentation Using Modulation Features, Generative Models, and Weighted Curve Evolution

In this work we approach the analysis and segmentation of natural textured images by combining ideas from image analysis and probabilistic modeling. We rely on AM-FM texture models and specifically on the Dominant Component Analysis (DCA) paradigm for feature extraction. This method provides a low-dimensional, dense and smooth descriptor, capturing essential aspects of texture, namely scale, orientation, and contrast. Our contributions are at three levels of the texture analysis and segmentation problems: First, at the feature extraction stage we propose a Regularized Demodulation Algorithm that provides more robust texture features and explore the merits of modifying the channel selection criterion of DCA. Second, we propose a probabilistic interpretation of DCA and Gabor filtering in general, in terms of Local Generative Models. Extending this point of view to edge detection facilitates the estimation of posterior probabilities for the edge and texture classes. Third, we propose the Weighted Curve Evolution scheme that enhances the Region Competition/ Geodesic Active Regions methods by allowing for the locally adaptive fusion of heterogeneous cues. Our segmentation results are evaluated on the Berkeley Segmentation Benchmark, and compare favorably to current state-of-the-art methods.

An iterative multi-scale tensor voting scheme for perceptual grouping of natural shapes in cluttered backgrounds

Grouping processes, which “organize” a given data by eliminating the irrelevant items and sorting the rest into groups, each corresponding to a particular object, can provide reliable pre-processed information to higher level computer vision functions, such as object detection and recognition. In this paper, we consider the problem of grouping oriented segments in highly cluttered images. In this context, we have developed a general and powerful method based on an iterative, multiscale tensor voting approach. Segments are represented as second-order tensors and communicate with each other through a voting scheme that incorporates the Gestalt principles of visual perception. The key idea of our approach is removing background segments conservatively on an iterative fashion, using multi-scale analysis, and re-voting on the retained segments. We have performed extensive experiments to evaluate the strengths and weaknesses of our approach using both synthetic and real images from publicly available datasets including the Williams and Thornber’s fruit-texture dataset [L. Williams, Fruit and texture images. Available from: <http://www.cs.unm.edu/~williams/saliency.html>, 2008 (last viewed in July 2008)] and the Berkeley segmentation dataset [C.F.P. Arbelaez, D. Martin, The berkeley segmentation dataset and benchmark. Available from: <http://www.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/segbench/>, 2008 (last viewed in July 2008)]. Our results and comparisons indicate that the proposed method improves segmentation results considerably, especially under severe background clutter. In particular, we show that using the iterative multiscale tensor voting approach to post-process the posterior probability map, produced by segmentation methods, improves boundary detection results in 84% of the grayscale test images in the Berkeley segmentation benchmark.

A customized Gabor filter for unsupervised color image segmentation

This paper presents work on accurate image segmentation utilizing local image characteristics. Image features are measured by employing an appropriate Gabor filter with adaptively chosen size, orientation, frequency and phase for each pixel. An image property called phase divergence is used for the selection of the appropriate filter size. Characteristic features related to the change in brightness, color, texture and position are extracted for each pixel at the selected size of the filter. In order to cluster the pixels into different regions, the joint distribution of these pixel features is modeled by a mixture of Gaussians utilizing three variants of the expectation maximization (EM) algorithm. The three different versions of EM used in this work for unsupervised clustering are: (1) penalized EM, (2) penalized stochastic EM, and (3) penalized inverse EM. Given the desired number of Gaussian mixture components, all three EM algorithms estimate the parameters of the mixture of Gaussians model that represents the joint distribution of pixel features. We determine the value of the number of models that best suits the natural number of clusters present in the image based on the Schwarz criterion, which maximizes the posterior probability of the number of groups given the samples of observation. This segmentation algorithm has been tested on the images of the Berkeley segmentation benchmark and the performance has demonstrated the effectiveness, accuracy and superiority of the proposed method.

Unsupervised multiscale color image segmentation based on MDL principle

We present an unsupervised multiscale color image segmentation algorithm. The basic idea is to apply mean shift clustering to obtain an over-segmentation and then merge regions at multiple scales to minimize the minimum description length criterion. The performance on the Berkeley segmentation benchmark campares favorably with some existing approaches.