Elsevier

Pattern Recognition

Volume 44, Issue 4, April 2011, Pages 777-787
Pattern Recognition

Color image segmentation using pixel wise support vector machine classification

https://doi.org/10.1016/j.patcog.2010.08.008Get rights and content

Abstract

Image segmentation is an important tool in image processing and can serve as an efficient front end to sophisticated algorithms and thereby simplify subsequent processing. In this paper, we present a color image segmentation using pixel wise support vector machine (SVM) classification. Firstly, the pixel-level color feature and texture feature of the image, which is used as input of SVM model (classifier), are extracted via the local homogeneity model and Gabor filter. Then, the SVM model (classifier) is trained by using FCM with the extracted pixel-level features. Finally, the color image is segmented with the trained SVM model (classifier). This image segmentation not only can fully take advantage of the local information of color image, but also the ability of SVM classifier. Experimental evidence shows that the proposed method has a very effective segmentation results and computational behavior, and decreases the time and increases the quality of color image segmentation in comparison with the state-of-the-art segmentation methods recently proposed in the literature.

Introduction

Image segmentation is a classic inverse problem, which consists of achieving a compact region-based description of the image scene by decomposing it into meaningful or spatially coherent regions sharing similar attributes. This low-level vision task is often the preliminary (and also crucial) step in many video and computer vision applications, such as object localization or recognition, data compression, tracking, image retrieval, or understanding. In recent years, a number of very inspiring and pioneering image segmentation algorithms have been developed, and these algorithms can be roughly classified into five major categories [1], [2]: thresholding [3], [4], template matching [5], [6], clustering [7], [8], [9], edge detection [10], [11], [12] and region growing [13], [14], [15]. These algorithms have been proven to be successful in many applications, but none of them are generally applicable to all images and different algorithms are usually not equally suitable for a particular application.

Image thresholding methods are popular due to their simplicity and efficiency. However, traditional histogram-based thresholding algorithms cannot separate those areas which have the same gray level but do not belong to the same part. In addition, they cannot process images whose histograms are nearly unimodal, especially when the target region is much smaller than the background area. Template matching method becomes time consuming when the image becomes more complex or larger in size. Clustering method, viewing an image as a set of multi-dimensional data and classifying the image into different parts according to certain homogeneity criterion, can get much better results of segmentation. But over-segmentation is the problem that must be settled and feature extraction is an important factor for clustering. The edge detection method is one of the widely used approaches to the problem of image segmentation. It is based on the detection of points with abrupt changes at gray levels. The main disadvantages of the edge detection technique are that it does not work well when images have many edges, and it cannot easily identify a closed curve or boundary. Region growing algorithms deal with spatial repartition of the image feature information. In general, they perform better than the thresholding approaches for several sets of images. However, the typical region growing processes are inherently sequential. The regions produced depend both on the order in which pixels are scanned and on the value of pixels which are first scanned and gathered to define each new segment. In view of the problems mentioned above, plenty of approaches and their corresponding improvements have been proposed to ensure the accuracy and rapidity of image segmentation. But there is still much work to be done to overcome their drawbacks, and attempts at utilizing knowledge on other domains, especially artificial intelligence, should be highly appreciated [2].

Recently, intelligent approaches, such as neural network and support vector machine (SVM) [16], have already been utilized successfully in image segmentation. Quan and Wen [17] proposed an effective multiscale method for the segmentation of the synthetic aperture radar (SAR) images via probabilistic neural network. By combining the probabilistic neural network (PNN) with the multiscale autoregressive (MAR) model, a classifier, which inherits the excellent strongpoint from both of them, is designed. Yu and Chang [18] presented an effective and efficient method for solving scenery image segmentation by applying the SVMs methodology. In scheme [19], the problem of scarcity of labeled pixels, required for segmentation of remotely sensed satellite images in supervised pixel classification framework, is addressed. Yan and Zheng [20] proposed a SAR image segmentation method based on one-class SVM, in which one-class SVM and two-class SVM for segmentation are discussed. Cyganek [21] proposed an efficient color segmentation method which is based on the SVM classifier operating in a one-class mode, and the method has been developed especially for the road signs recognition system. In scheme [22], support vector clustering (SVC) is used for marketing segmentation, and a case study of a drink company is used to demonstrate the proposed method and compared with the k-means and the self-organizing feature map (SOFM) methods. Ji et al. [23] proposed a new semi-supervised approach based on transductive support vector machine (TSVM) to segment SAR images, and it is robust to noises and is effective when dealing with low numbers of high-dimensional samples. Xue et al. [24] presented a method of automatic segmentation and classification of mosaic patterns in cervigrams in which a support vector machine (SVM) classifier is applied, using learning from a “ground truth” dataset annotated by medical experts in oncology and gynecology. In the approach, the acetowhite region is split into tiles, and texture features are extracted from each tile. The SVM classifier is trained using the texture features of tiles obtained from ground truth images. Given a new test image, the trained SVM classifier is applied to classify each tile in the test image, and the classified tiles are combined to generate the final segmentation map.

In this paper, we propose a color image segmentation using pixel wise support vector machine (SVM) classification. Firstly, the pixel-level color feature and texture feature of the image, which is used as input of SVM model (classifier), are extracted via the local homogeneity model and Gabor filter. Then, the SVM model (classifier) is trained by using FCM with the extracted pixel-level features. Finally, the color image is segmented with the trained SVM model (classifier). This image segmentation not only can fully take advantage of the local information of color image, but also the ability of SVM classifier. Simulation results show that the proposed method achieves competitive segmentation results compared to the state-of-the-art segmentation methods recently proposed in the literature.

The rest of this paper is organized as follows. Section 2 presents the basic theory about SVM. In Section 3, the pixel-level color feature and texture feature extraction are described. Section 4 contains the description of our color image segmentation. Simulation results in Section 5 will show the performance of our scheme. Finally, Section 6 concludes this presentation.

Section snippets

The support vector machine (SVM)

Support vector machines (SVM) have been successfully applied in classification and function estimation problems after their introduction by Vapnik within the context of statistical learning theory and structural risk minimization [25]. Vapnik constructed the standard SVM to separate training data into two classes. The goal of the SVM is to find the hyper-plane that maximizes the minimum distance between any data point, as shown in Fig. 1.

Given a training dataset of l points {xi,yi}i=1l with the

The pixel-level color and texture feature

In this paper, each pixel of an image is identified as belonging to a homogenous region corresponding to an object or part of an object. The problem of image segmentation is regarded as a classification task, and the goal of segmentation is to assign a label to individual pixel or a region. So, it is very important to extract the effective pixel-level image feature. Here, we extract the pixel-level color and texture feature via the local homogeneity model and Gabor filter.

The pixel-based color image segmentation using SVM and FCM

We know that the image segmentation can be taken as classification problems, which can be solved using anyone of well-known classification techniques. SVM is one of the classification techniques and good results of the SVM technique in pattern recognition have been obtained, so we can choose the SVM for solving color image segmentation problems. In this paper, we present a pixel-based color image segmentation using SVM and FCM. Firstly, the pixel-level color feature and texture feature of the

Evaluation setup and dataset

Comprehensive experiments were conducted in natural scene images to evaluate the performance of our image segmentation method. The proposed method has been used to segment an image into distinct color-textured regions on the Berkeley segmentation database [1]. This database was selected because it contains hand-labeled segmentations of the images from 30 human subjects. Half of the segmentations involve color images and the other half grayscale images. The database comprises of various images

Conclusions

Image segmentation is an important low-level preprocessing step for many computer vision problems. In this paper, we have presented a new approach for color image segmentation based on SVM and FCM. Firstly, the pixel-level color feature and texture feature of the image, which is used as input of SVM model (classifier), are extracted via the local homogeneity model and Gabor filter. Then, the SVM model (classifier) is trained by using FCM with the extracted pixel-level features. Finally, the

Acknowledgements

This work was supported by the National Natural Science Foundation of China under Grant nos. 60773031 and 60873222, the Open Foundation of State Key Laboratory of Networking and Switching Technology of China under Grant no. SKLNST-2008-1-01, the Open Foundation of State Key Laboratory of Information Security of China under Grant no. 03-06, the Open Foundation of State Key Laboratory for Novel Software Technology of China under Grant no. A200702, and Liaoning Research Project for Institutions of

Xiangyang Wang was born in Tieling, China, in 1965. He is currently a professor with the School of Computer and Information Technology at the Liaoning Normal University, China. He obtained his B.S. degree from the Lanzhou University, China and his M.S. degree from the Jilin University, China, in 1988 and 1995, respectively. His research interests include signal processing and communications, digital multimedia data hiding and information assurance, applications of digital image processing,

References (27)

  • X. Wang et al.

    A color- and texture-based image segmentation algorithm

    Machine Graphics & Vision

    (2010)
  • J. Shi et al.

    Normalized cuts and image segmentation

    IEEE Transactions on Pattern Analysis and Machine Intelligence

    (2000)
  • P. Bao et al.

    Canny edge detection enhancement by scale multiplication

    IEEE Transactions on Pattern Analysis and Machine Intelligence

    (2005)
  • Cited by (150)

    • Hyperspectral remote sensing image classification based on tighter random projection with minimal intra-class variance algorithm

      2021, Pattern Recognition
      Citation Excerpt :

      It can be divided into supervised classification algorithm [25] and unsupervised classification algorithm [26] according to the condition of prior information. The representative supervised classification algorithms include Minimum Distance (MD) [27], Convolution Neural Network (CNN) [28], and Support Vector Machine (SVM) [29]. Su et al. [30] proposed a land parcel extraction algorithm based on training data and MD classifier, which combines PCA for extracting the first principal component of the original satellite image, MD for pre-classification, and watershed algorithm for subdividing.

    View all citing articles on Scopus

    Xiangyang Wang was born in Tieling, China, in 1965. He is currently a professor with the School of Computer and Information Technology at the Liaoning Normal University, China. He obtained his B.S. degree from the Lanzhou University, China and his M.S. degree from the Jilin University, China, in 1988 and 1995, respectively. His research interests include signal processing and communications, digital multimedia data hiding and information assurance, applications of digital image processing, computer vision. He has published more than 150 journal papers, 20 conference papers, and contributed in 2 books in his areas of interest.

    Ting Wang received the B.S. degree from the School of Computer and Information Technology, Liaoning Normal University, China, in 2008, where she is currently pursuing the M.S. degree. Her research interests include signal processing and image segmentation.

    Juan Bu received the B.S. degree from the School of Computer and Information Technology, Liaoning Normal University, China, in 2007, where she is currently pursuing the M.S. degree. Her research interests include signal processing and image segmentation.

    View full text