Hybrid Soft Computing for Image Segmentation

This chapter proposed multilevel thresholding hybrid swarms optimization algorithm for image segmentation. The proposed algorithm is inspired by the behavior of fireflies and real spider. It uses Firefly Algorithm (FA) and Social Spider Optimization (SSO) algorithm (FASSO). The objective function used for achieving multilevel thresholding is the maximum between class variance criterion. The proposed algorithm uses the FA to optimize threshold, and then uses this thresholding value to partition the images through SSO algorithm of a powerful global search capability. Experimental results demonstrate the effectiveness of the FASSO algorithm of image segmentation and provide faster convergence with relatively lower CPU time.

Multilevel image thresholding plays a crucial role in analyzing and interpreting the digital images. Previous studies revealed that classical exhaustive search techniques are time consuming as the number of thresholds increased. To solve the problem, many nature-inspired algorithms (NAs) which can produce high-quality solutions in reasonable time have been utilized for multilevel thresholding. This chapter discusses three typical kinds of NAs and their hybridizations in solving multilevel image thresholding. Accordingly, a novel hybrid algorithm of gravitational search algorithm (GSA) with genetic algorithm (GA), named GSA-GA, is proposed to explore optimal threshold values efficiently. The chosen objective functions in this chapter are Kapur’s entropy and Otsu criteria. This chapter conducted experiments on two well-known test images and two real satellite images using various numbers of thresholds to evaluate the performance of different NAs.

Thresholding is the key to simplify image classification. It becomes challenging when the number of thresholds is more than two. Most of the existing multilevel thresholding techniques use image histogram information (first-order statistics). This chapter utilizes optimal edge magnitude information (second-order statistics) of an image to obtain multilevel threshold values. We compute the edge magnitude information from the gray-level co-occurrence matrix (GLCM) of the image. The second-order statistics uses the correlation among the pixels for improved results. Maximization of edge magnitude is vital for obtaining optimal threshold values. The edge magnitude is maximized by introducing a novel hybrid cuckoo search-bacterial foraging optimization (CS-BFO) algorithm. The novelty of our proposed CS-BFO algorithm lies in its ability to provide improved chemotaxis in BFO algorithm, which is achieved by supplementing levy flight feature of CS. Social foraging models are relatively efficient for determining optimum multilevel threshold values. Hence, CS-BFO is used for improved thresholding performance and highlighting the novelty of this contribution. We have also implemented other soft computing tools cuckoo search (CS), particle swarm optimization (PSO), and genetic algorithm (GA) for a comparison. In addition, we have incorporated constraint handling in all the above-mentioned techniques so that optimal threshold values do not cross the bounds. This study reveals the fact that CS technique provides us improved speed while the CS-BFO method shows improved results both qualitatively and quantitatively.

Image segmentation as a concept has great potential for practical implementation. Image segmentation is complex concept, which can be focused on object recognition within images, or content similarity oriented concept, or some other concept which in general try to recognize similar elements within images. This chapter will introduce novel concept based on picture content similarity, which can be used as a tool for recommendation systems in situation where we operate with unknown image data sets. Simple example for that can be travel agencies web pages, where potential users are seeking for future travel destinations. Based on experience regarding previously visited locations, system calculates content similarity between mostly visited locations represented by pictures and offers new locations upon calculated preference. Generally speaking, this process can be declared as image segmentation, because segmentation is based on example picture content, and all locations similar to the chosen picture content are declared as segment. Visitor gets recommendation for his next travel destination based on previously seen locations within one web session, or more than one web session if visitor can be uniquely recognized by login and password. Method presented in this chapter should be a good solution for online systems which demands fastness and efficiency for recommendation systems. Existing methods are much precise, but demands longest computing processing. Proposed method does not demand exhaustive training to be efficient in image classification. Partially it cause with less precision in comparison with methods such as histogram-based methods, compression-based methods, algorithms based on edge detection, region growing, partial differential equation-based methods, and others, but speed up process of image classification.

An application of hybrid soft computing technique for early detection and treatment of a most common mosquito-borne viral disease Dengue, is discussed thoroughly in this chapter. The global pictures of dengue endemics are shown clearly. The structure of dengue virus and the infection procedure of the virus are also discussed. A detailed analysis of dengue illness, diagnosis methods, and treatments has been done to conclude that platelet counting is needful for early diagnosis of Dengue illness and for monitoring the health status of the patients. The main challenge in developing an automated platelet counting system for efficient, easy, and fast detection of dengue infection as well as treatment, is in the segmentation of platelets from microscopic images of a blood smear. This chapter shows how the challenges can be overcome. Color-based segmentation and k-means clustering cannot provide desired outputs in all possible situations. A hybrid soft computing technique efficiently segments platelet and overcomes the shortcomings of the other two segmentation techniques. This technique is the combination of fuzzy c-means technique and adaptive network-based fuzzy interference system (ANFIS). We have applied three different segmentation techniques namely color-based segmentation, k-means, and the hybrid soft computing technique on poor intensity images. However, only the hybrid soft computing technique detects the platelets correctly.

The processing of ocean satellite images has as goal the detection of phenomena related with ocean dynamics. In this context, Mesoscale Oceanic Structures (MOS) play an essential role. In this chapter we will present the tool developed in our group in order to extract knowledge rules for the retrieval of MOS in ocean satellite images. We will describe the implementation of the tool: the workflow associated with the tool, the user interface, the class structure, and the database of the tool. Additionally, the experimental results obtained with the tool in terms of fuzzy knowledge rules as well as labeled structures with these rules are shown. These results have been obtained with the tool analyzing chlorophyll and temperature images of the Canary Islands and North West African coast captured by the SeaWiFS and MODIS-Aqua sensors.

Segmentation of an image is an essential assignment in an image examination whereby picture is divided into significant areas whose focuses have almost the same properties like; dim levels, mean qualities, or text-related characteristics. The pictures are divided into locales which best speak to the important articles in the scene. Locale parameters, for example, territory, shape, measurable parameters, and surface can be extricated and utilized for further examination of information. The examination of satellite symbolism of common scenes presents numerous one of a kind issues and it varies from an investigation and division of urban, business, or agrarian ranges. Once the division classes of a picture is obtained, it is conceivable to utilize heuristics or other area particular ways to deal with further characterize, translate, comprehend, register or extract information from the partitioned image. The applications of analysis of satellite imagery is plenty in real-life situations like weather forecasting, analysis of natural scenes, urban planning, environmental monitoring, object recognition, detection of mass wasting, etc. are well known. Several algorithms using classical approaches as well as those using uncertainty-based approaches have been proposed. The analysis shows that hybrid approaches are more efficient than the individual ones. In this chapter, we discuss on all the uncertainty-based and hybrid algorithms for segmentation of satellite imagery and their applications. Also, we propose some open problems which can be handled for future work.

Human skin segmentation has several applications in computer vision beyond its main purpose of distinguishing between skin and nonskin regions. Despite the large number of methods available in the literature, accurate skin segmentation is still a challenging task. Many methods rely only on color information, which does not completely discriminate the image regions due to variations in lighting conditions and ambiguity between skin and background color. This chapter extends upon a self-contained method for skin segmentation that outlines regions from which the overall skin color can be estimated and such that the color model is adjusted to a particular image. This process is based on thresholds that were empirically defined in a first approach. The proposed method has three main contributions over the previous one. First, genetic algorithm (GA) is applied to search for better thresholds that will be used to extract appropriate seeds from the general probability and texture maps. Next, the GA is also applied to define thresholds for edge detectors aiming to improve edge connections. Finally, a fuzzy method for fusion is included where its parameters are optimized by GA during a learning phase. The improvements added to the skin segmentation method are evaluated on a set of hand gesture images. A statistical analysis is conducted over the computational results achieved by each evaluated method, indicating a superior performance of our novel skin segmentation method.

Data clustering has been an integral and important part of data mining. It has wide applications in database anonymization, decision making, image processing and pattern recognition, medical diagnosis, and geographical information systems, only to name a few. Data in real-life scenario are having imprecision inherent in them. So, early crisp clustering techniques are very less efficient. Several imprecision-based models have been proposed over the years. Of late, it has been established that the hybrid models obtained as combination of these imprecise models are far more efficient than the individual ones. Several clustering algorithms have been put forth using these hybrid models. It is also found that conventional fuzzy clustering algorithms fail in incorporating the spatial information. This chapter focuses on discussing some of the spatial data clustering algorithms developed so far and their applications mainly in the area of image segmentation.

This paper presents a novel method based on single-scale Gabor filters (SSG) consisting of three steps for vessel segmentation and vessel width estimation of X-ray coronary angiograms. In the first stage, a comparative analysis of genetic algorithms, and two estimation of distribution algorithms in order to improve the vessel detection rate of the SSG, while reducing the computational time of the training step is performed. The detection results of the SSG are compared with those obtained by four state-of-the-art detection methods via the area (\(A_z\)) under the receiver operating characteristic (ROC) curve. In the second stage, a comparative analysis of five automatic thresholding methods is performed in order to discriminate vessel and nonvessel pixels from the Gabor filter response. In the last step, a procedure to estimate the vessel width of the segmented coronary tree structure is presented. The experimental results using the SSG obtained the highest vessel detection performance with \(A_z = 0.9584\) with a training set of 40 angiograms. In addition, the segmentation results using the interclass variance thresholding method provided a segmentation accuracy of 0.941 with a test set of 40 angiograms. The performance of the proposed method consisting of the steps of vessel detection, segmentation, and vessel width estimation shows promising results according to the evaluation measures, which is suitable for clinical decision support in cardiology.

The incidence of breast cancer has rapidly increased over the past few decades in India and the mortality rate is more than other countries across the entire world. These facts have motivated the development of new technologies or modification of the existing technologies for the identification of breast cancer before it metastasizes to the neighboring tissues. Breast thermography is a promising front-line breast screening method, which is noncontact, cheap, quick, economic, and painless. The use of thermal imaging for the identification of breast abnormality is based on the principle that the temperature distribution in precancerous tissue and its surrounding area are always higher than that in normal breast tissue. However, the accurate interpretation and classification of the breast thermograms for proper diagnostic decision-making is a major problem. Proper segmentation of hottest region from the segmented breast region plays a key part in the diagnosis of breast cancer that calls for the application of hybrid intelligent methods in the segmentation of hottest region. The shape and size of the hottest regions are used to determine the degree of malignancy of the tumor and classify its type. Hybrid intelligent systems have been successfully applied in the classification of breast thermal images over the last few years. In this chapter, we have proposed a sequential hybrid intelligent technique for the segmentation of the hottest region and also shown the significance of hybrid intelligence systems over the conventional methods for the segmentation of hottest region. A detailed review related to the segmentation of breast region and the segmentation of hottest region is included in this chapter. In addition, this chapter also contains the detailed overview of the principles, reliability, and predictive ability of the breast thermogram in early diagnosis of breast cancer.

Probabilistic Features Combination method (PFC), is the approach of multidimensional data modeling, extrapolation and interpolation using the set of high-dimensional feature vectors. This method is a hybridization of numerical methods and probabilistic methods with N-dimensional data interpolation for feature vectors. Each feature is treated as a random variable.