Differential Evolutionary Superpixel Segmentation

Combining contour-based and region-based in image segmentation

Background

This paper presents an optimized clustering approach applied to image segmentation. Accurate image segmentation impacts many fields like medical, machine vision, object detection. Applications involve tumor detection, face detection and recognition, and video surveillance.

Methods

The developed approach is based on obtaining an optimum number of clusters and regions of an image. We combined Region-based and contour-based approaches. Initial rough regions are obtained using edge detection. We have used Gabor wavelets for texture classification and spatial resolutions. Color frequencies are also used to determine the number of clusters of the Fuzzy c-means (FCM) algorithm which gave an optimum number of clusters or regions.

Results

We have compared our approach with other similar wavelet and clustering techniques. Our algorithm gave better values for segmentation metrics like SNR, PSNR, and MCC.

Conclusions

Optimizing the number of clusters or regions has a significant effect on the performance of the image segmentation techniques. This will result in better detection and localization of the segmentation-based application.

Deep learning approach based on superpixel segmentation assisted labeling for automatic pressure ulcer diagnosis

A pressure ulcer is an injury of the skin and underlying tissues adjacent to a bony eminence. Patients who suffer from this disease may have difficulty accessing medical care. Recently, the COVID-19 pandemic has exacerbated this situation. Automatic diagnosis based on machine learning (ML) brings promising solutions. Traditional ML requires complicated preprocessing steps for feature extraction. Its clinical applications are thus limited to particular datasets. Deep learning (DL), which extracts features from convolution layers, can embrace larger datasets that might be deliberately excluded in traditional algorithms. However, DL requires large sets of domain specific labeled data for training. Labeling various tissues of pressure ulcers is a challenge even for experienced plastic surgeons. We propose a superpixel-assisted, region-based method of labeling images for tissue classification. The boundary-based method is applied to create a dataset for wound and re-epithelialization (re-ep) segmentation. Five popular DL models (U-Net, DeeplabV3, PsPNet, FPN, and Mask R-CNN) with encoder (ResNet-101) were trained on the two datasets. A total of 2836 images of pressure ulcers were labeled for tissue classification, while 2893 images were labeled for wound and re-ep segmentation. All five models had satisfactory results. DeeplabV3 had the best performance on both tasks with a precision of 0.9915, recall of 0.9915 and accuracy of 0.9957 on the tissue classification; and a precision of 0.9888, recall of 0.9887 and accuracy of 0.9925 on the wound and re-ep segmentation task. Combining segmentation results with clinical data, our algorithm can detect the signs of wound healing, monitor the progress of healing, estimate the wound size, and suggest the need for surgical debridement.

A Reference Point Selection and Direction Guidance-Based Algorithm for Large-Scale Multi-Objective Optimization

Designing efficient algorithms for the large-scale multi-objective problems (LSMOPs) is a very challenging problem currently. To tackle LSMOPs, we first design a new reference points selection strategy to enhance the diversity of algorithms and avoid the algorithm falling into local minima. The strategy selects not only a part of nondominated solutions with the largest crowding distance, but also a part of relatively uniformly distributed solutions as the reference points. In this way, much better diversity and convergence can be obtained. Second, we propose a direction-guided offspring generation strategy, where a type of potential directions is designed to generate the promising solutions which can balance the convergence and diversity of the obtained solutions and improve the effectiveness of the algorithm significantly. Based on the two proposed strategies, we propose a new effective algorithm for LSMOPs. Numerical experiments are executed on two widely used large-scale multi-objective benchmark problem sets with 200, 500 and 1000 decision variables and a comparison with five state-of-the-art algorithms is made. The experimental results show that our proposed algorithm is effective and can obtain significantly better solutions than the compared algorithms.

A Two-Phase Learning-Based Swarm Optimizer for Large-Scale Optimization

In this article, a simple yet effective method, called a two-phase learning-based swarm optimizer (TPLSO), is proposed for large-scale optimization. Inspired by the cooperative learning behavior in human society, mass learning and elite learning are involved in TPLSO. In the mass learning phase, TPLSO randomly selects three particles to form a study group and then adopts a competitive mechanism to update the members of the study group. Then, we sort all of the particles in the swarm and pick out the elite particles that have better fitness values. In the elite learning phase, the elite particles learn from each other to further search for more promising areas. The theoretical analysis of TPLSO exploration and exploitation abilities is performed and compared with several popular particle swarm optimizers. Comparative experiments on two widely used large-scale benchmark datasets demonstrate that the proposed TPLSO achieves better performance on diverse large-scale problems than several state-of-the-art algorithms.

Multilevel Image Thresholding Based on Improved Expectation Maximization (EM) and Differential Evolution Algorithm

Multilevel image thresholding is an essential step in the image segmentation process. Expectation Maximization (EM) is a powerful technique to find thresholds but is sensitive to the initial points. Differential Evolution (DE) is a robust metaheuristic algorithm that can find thresholds rapidly. However, it may be trapped in the local optimums and premature convergence occurs. In this paper, we incorporate EM algorithm to DE and introduce a novel algorithm called EM+DE which overcomes these shortages and can segment images better than EM and DE algorithms. In the proposed method, EM estimates Gaussian Mixture Model (GMM) coefficients of the histogram and DE tries to provide good volunteer solutions to EM algorithm when EM converges in local areas. Finally, DE fits GMM parameters based on Root Mean Square Error (RMSE) to reach the fittest curve. Ten standard test images and six famous metaheuristic algorithms are considered and result on global fitness. PSNR, SSIM, FSIM criteria and the computational time are given. The experimental results prove that the proposed algorithm outperforms the EM and DE as well as EM+ other natural-inspired algorithms in terms of segmentation criteria.

A Darwinian Differential Evolution Algorithm for Multilevel Image Thresholding

Image segmentation is a prime operation to understand the content of images. Multilevel thresholding is applied in image segmentation because of its speed and accuracy. In this paper, a novel multilevel thresholding algorithm based on differential evolution (DE) search is introduced. One of the major drawbacks of metaheuristic algorithms is the stagnation phenomenon which leads to falling into local optimums and premature convergence. To overcome this shortcoming, the idea of Darwinian theory is incorporated with DE algorithm to increase the diversity and quality of the individuals without decreasing the convergence speed of DE algorithm. A policy of encouragement and punishment is considered to lead searching agents in the search space and reduce the computational time. The algorithm is implemented based on dividing the population into specified groups and each group tries to find a better location. Ten test images are selected to verify the ability of our algorithm using the famous energy curve method. Kapur entropy and Type 2 fuzzy entropy are employed to evaluate the capability of the introduced algorithm. Nine different metaheuristic algorithms with Darwinian modes are also implemented and compared with our method. Experimental results manifest that the proposed method is a powerful tool for multilevel thresholding and the obtained results outperform the DE algorithm and other methods.

High-Resolution Rice Mapping Based on SNIC Segmentation and Multi-Source Remote Sensing Images

High-resolution crop mapping is of great significance in agricultural monitoring, precision agriculture, and providing critical information for crop yield or disaster monitoring. Meanwhile, medium resolution time-series optical and synthetic aperture radar (SAR) images can provide useful phenological information. Combining high-resolution satellite data and medium resolution time-series images provides a great opportunity for fine crop mapping. Simple Non-Iterative Clustering (SNIC) is a state-of-the-art image segmentation algorithm that shows the advantages of efficiency and high accuracy. However, the application of SNIC in crop mapping based on the combination of high-resolution and medium-resolution images is unknown. Besides, there is still little research on the influence of the superpixel size (one of the key user-defined parameters of the SNIC method) on classification accuracy. In this study, we employed a 2 m high-resolution GF-1 pan-sharpened image and 10 m medium resolution time-series Sentinel-1 C-band Synthetic Aperture Radar Instrument (C-SAR) and Sentinel-2 Multispectral Instrument (MSI) images to carry out rice mapping based on the SNIC method. The results show that with the increase of the superpixel size, the classification accuracy increased at first and then decreased rapidly after reaching the summit when the superpixel size is 27. The classification accuracy of the combined use of optical and SAR data is higher than that using only Sentinel-2 MSI or Sentinel-1 C-SAR vertical transmitted and vertical received (VV) or vertical transmitted and horizontal received (VH) data, with overall accuracies of 0.8335, 0.8282, 0.7862, and 0.7886, respectively. Meanwhile, the results also indicate that classification based on superpixels obtained by SNIC significantly outperforms classification based on original pixels. The overall accuracy, producer accuracy, and user accuracy of SNIC superpixel-based classification increased by 9.14%, 17.16%, 27.35% and 1.36%, respectively, when compared with the pixel-based classification, based on the combination of optical and SAR data (using the random forest as the classifier). The results show that SNIC superpixel segmentation is a feasible method for high-resolution crop mapping based on multi-source remote sensing data. The automatic selection of the optimal superpixel size of SNIC will be focused on in future research.

RGB-'D' Saliency Detection With Pseudo Depth

Recent studies have shown the effectiveness of using depth information in salient object detection. However, the most commonly seen images so far are still RGB images that do not contain the depth data. Meanwhile, the human brain can extract the geometric model of a scene from an RGB-only image and hence provides a 3D perception of the scene. Inspired by this observation, we propose a new concept named RGB-'D' saliency detection, which derives pseudo depth from the RGB images and then performs 3D saliency detection. The pseudo depth can be utilized as image features, prior knowledge, an additional image channel, or independent depth-induced models to boost the performance of traditional RGB saliency models. As an illustration, we develop a new salient object detection algorithm that uses the pseudo depth to derive a depth-driven background prior and a depth contrast feature. Extensive experiments on several standard databases validate the promising performance of the proposed algorithm. In addition, we also adapt two supervised RGB saliency models to our RGB-'D' saliency framework for performance enhancement. The results further demonstrate the generalization ability of the proposed RGB-'D' saliency framework.

Content-Adaptive Superpixel Segmentation

Superpixel segmentation targets at grouping pixels in an image into atomic regions whose boundaries align well with the natural object boundaries. This paper first proposes a new feature representation for superpixel segmentation that holistically embraces color, contour, texture, and spatial features. Then, we introduce a clustering-based discriminability measure to iteratively evaluate the importance of different features. Integrating the feature representation and the discriminability measure, we propose a novel content-adaptive superpixel (CAS) segmentation algorithm. CAS is able to automatically and iteratively adjust the weights of different features to fit various properties of image instances. Experiments on several challenging datasets demonstrate that the proposed CAS outperforms the state-of-the-art methods and has a low computational cost.