Unsupervised image segmentation (UIS) is a very challenging problem in computer vision and artificial intelligence. In general, the UIS can be viewed as a clustering problem, which aims to partition the image into K sets of regions that have coherent color and texture features. Clearly, the UIS results depend on the employed clustering methods. Inspired by the recent rapid progress of l 1 - norm minimization techniques and the great success of l 1 - norm minimization based sparse representation (SR), in this paper we propose a K -sparse clustering algorithm to segment the image into K partitions. The given image is first over-segmented into many small regions, from which the K clustering centers are learned under the SR framework using l 1 - norm minimization. Finally, each of the regions can be classified into one of the K classes based on its SR coefficients over the K cluster centers. The experimental results on the benchmark images demonstrate the effectiveness and superiority of the proposed K -sparse clustering method to the classical K -means clustering method and recently developed sparse subspace clustering method in applications of UIS.

A new algorithm for image segmentation based on k-mean and particle swarm optimization algorithm is proposed in the paper. K-mean clustering algorithm is a local search algorithm because it is easily trapped local optimum and is sensitive to initial value effectively. On the other hand, particle swarm optimization algorithm is a global optimization algorithm. By incorporating the local search ability of k-mean algorithm and the global optimization ability of PSO and taking the criterion function of k-mean as the object function of PSO, a new hybrid color image segmentation algorithm based on particle swarm optimization and k-mean algorithm is proposed. Experiments show that the new algorithm can get the optimal quantizated image by PSNR and RMSE.

In this paper, we proposed the wavelet transform and color space object segmentation. Image segmentation for image brightness has a great influence, it is necessary to adjust the brightness of the image, and select the correct color space and definition of success threshold to make a more complete image can be divided. The wavelet transform can not only compress image, but also improve the lighting problem, we use improving variable illumination on wavelet(IVIW) to adjust the brightness, then after several wavelet image decomposition to obtained by the average gray scale image of pixel, making the image brightness normalize and reconstruction. There are several types of color space, due to Ycbcr color space have good response of light, we use the RGB color convert to the Ycbcr color space with extract the luminance to threshold segmentation. Experimental results show that the method effectively improved the quality of image segmentation.

ESA is a new method of word segmentation presented in our early paper. Because the whole method has only one parameter to be estimated, it is the most unsupervised method for the task to date. More importantly, the parameter can be approximately predicted by the empirical formulae based on SIGHAN Bakeoff-2 dataset. In this article, we further evaluate the empirical formulae on SIGHAN Bakeoff-3 dataset. The evaluation results show that the original formulae provide quite accurate prediction of the parameter. Finally, we correct the formulae according to both datasets for the better prediction.

Segmentation of medical images, particularly magnetic resonance images of brain is complex and it is considered as a huge challenge in image processing. Among the numerous algorithms presented in this context, the fuzzy Cmean (FCM) algorithm is widely used in MR images segmentation. Recently, researchers have introduced two new parameters in order to improve the performance of FCM algorithm, which are calculated using neural network in a complex and time consuming manner. These two parameters have been then calculated by other researchers using genetic algorithm (GA) and particle swarm optimization (PSO) algorithm, which although it has reduced the time but no change obtained in the resulted segmentation quality. In this paper we calculate these two parameters using the artificial bee colony (ABC) algorithm aiming to both reduce the time and to reach a higher quality than that obtained by previous reports.

In multimedia based learning, video is widely used in many course domains. Recently power point presentation is widely used in the educational institutes. We address the problem of the prerecorded lecture video organization of visual information through user's interaction at different steps. In current situation it is very difficult to browse the recorded video, since no information about the contents of the video are provided. NPTEL- Joint venture of IIT's and IIS's main objective is to enhance the quality of engineering education in the country by developing curriculum based video and web courses. This is being carried out by seven IITs and IISc Bangalore as a collaborative project. But videos present cannot be used effectively, because of its poor browsing facility. Video segmentation is done in 3 steps as follows. First get the pre recorded lecture video as input. The lecture video is broken into Frames. Then the important frames (slides) are selected for indexing using the contents of the frames and index points are retrieved. Finally Original prerecorded lecture video start from particular retrieved index point.

People counting system are used to improve attendance rates and security management in different areas. These areas such as museums, library, improve its receptions during peak hours thanks to this application. Public area such as malls, need to manage areas during peak hours too. In some circumstances, it is used to count the number of customer on one location for one specific moment, taking into consideration that there is a special event (discount in a store, manifestation, street festival). In this paper, we propose a framework to realize a good people counting process. The framework includes various steps such as: acquisition, motion detection, segmentation, tracking and counting steps. Experimental results are presented and discussed with different video input.

Using thresholding method to segment an image, a fixed threshold is not suitable if the background is rough Here, we propose a new adaptive thresholding method using FCM. The method requires only one parameter to be selected and the adaptive threshold surface can be found automatically from the original image.An adaptive thresholding scheme using adaptive tracking and morphological filtering. FCM algorithm computes the fuzzy membership values for each pixel. Our method is good for detecting large and small images concurrently. It is also efficient to denoise and enhance the responses of images with low local contrast can be detected. The efficiency and accuracy of the algorithm is demonstrated by the experiments on the MR brain images.

A novel automated Fuzzy Optimal Thresholding Technique to differentiate malignant melanoma lesions from the normal skin is proposed in this paper. Segmentation of pigmented skin lesion images pertaining to malignant melanoma diagnosis is achieved with the help of fuzzy Optimal Threshold value and Mathematical morphology. The input melanoma skin lesion color images are acquired and preprocessed to remove noise as well as some unwanted structures like hairs, etc. The preprocessed image is smoothened by morphological operations, converted into a gray scale image and binarized by using Fuzzy Optimal Threshold value obtained by Fuzzy Gaussian Measures. The largest label region, a cancer area from the binary image is identified and the borders are extracted and imposed on the original skin lesion image. The efficiency of the algorithm is tested over 100 images and proved up to 90% of the accurate extraction of skin lesion borders.

Image segmentation is a fundamental problem in image processing, the traditional watershed image segmentation algorithm has a simple, intuitive, and many other advantages, however, there are over-segmentation, great impact of noise and difficulty for image pre-processing problems etc. so that propose an improved algorithm based on morphological reconstruction, which can solve the noise problem and the reducing difficulty of pre-treatment and improve the segmentation results. In the traditional watershed algorithm, apply a faster sorting algorithm, a reasonable course of treatment uncertainties in flood, it is quick and effective for image segmentation. Experiments show that the algorithm can effectively eliminate the phenomenon of over-segmentation, and good to improve the computing speed and accuracy for watershed algorithm.

A line scan camera is used in an image based aggregate analysis. Willi this camera some algorithms are designed to segment stones from background. When the image acquisition is going, dust and some unwanted waste objects are mixed with stone material. They form a kind of stripped background in acquired images. The stripped background produced by the line scan camera scales up the difficulties of segmentation. To deal with such circumstance, a background removing algorithm is proposed as first stage, and then the first order statistic filter is used to improve result. Next stage is binarization in which the OTSU method is adopted. Finally the image is segmented using a sweep-line region growing algorithm. The algorithm just scans the image line by line from top to bottom while region growing is made from pixels in stones. The feature of the algorithm is the directions of growing are only the vertical and the horizontal. It makes the algorithm simple and effective. The processed pictures show pretty good results in every stage.

We have been engaged in the development of an effective English and Chinese bilingual web content categorization engine for some years. Due to the nature of the two languages, the processing algorithms for the two languages also differ significantly. In this paper, we evaluate a number of segmentation methods for Chinese text with the expressed purposes of analyzing web textual content information for effective web content filtering. Based on the evaluation results, a specific method is adapted for the task.

This paper discusses the application of Culture Algorithm to Fuzzy C-means clustering. The aim is to improve the quality of image segmentation. Fuzzy C-means clustering algorithm as one of the fuzzy clustering is widely used in image segmentation. We use Culture Algorithm to initialize centre of clusters to make Fuzzy C-means robust against initialization. The proposed method can effectively avoid being trapped into local optimum as well. The analysis has shown that Fuzzy Clustering based on Culture Algorithm can obtain better total clustering accuracy with high efficiency, and the quality of image segmentation is improved.

This paper presents self-localization module for mobile robot that travels around indoor environments. Our module uses the only one sensor, a single camera looking up the ceiling. There is no efficient enough SLAM (Simultaneous Localization And Mapping) algorithm working on embedded system. The initial difficulty of vision based SLAM is computational complexity to acquire reliable feature on their algorithm. To reduce the computational complexity, we use the ceiling segmentation to extract line features of ceiling area. Line features are extracted from the boundaries between the ceiling and walls. Extended Kalman Filter is used to estimate the pose of a robot and build the ceiling map with line features. The experiment is practiced in our indoor test bed and the proposed algorithm is proved by the experimental results.

This paper proposes an effective method for slap fingerprint segmentation using connected component analysis. First, the gray-scale image scanned from ten-print card is preprocessed with down-sampling and binarization. Then, connected components are found, labeled, analyzed and processed to obtain the candidate fingerprint images for each finger. Finally, each fingerprint image is carefully assigned to its corresponding finger. The experimental results show that proposed method can segment the slap fingerprints effectively and accurately.

Evolving Clustering Method is an effective method of image segmentation, but there are several defects in it. In this paper, we intend to use the Fuzzy C-Means algorithm to optimize the results of ECM algorithm. The experiments confirmed the optimization is valid.

This paper presents a hybrid method, low-resolution and high-resolution, for Persian page segmentation. In the low-resolution page segmentation, a pyramidal image structure is constructed for multi scale analysis and segments document image to a set of regions. By high-resolution page segmentation, by connected components analysis, each region is segmented to homogeneous regions and identifying them as texts, images, and tables/drawings. The proposed method was experiment with the Persian documents. The result of these tests have shown that the proposed method provide more accurate and speed results.

In this paper, we propose a novel method to estimate the haze-free image based on the improved optical model. In this work, we calculate the objective transmission and distance transmission to estimate the haze-free image by the segmented hazy image. Firstly, the color clustering method is used to segment the image into several regions by color similarity for getting the objective transmission. Then, the graph-based segmentation is used to calculate the depth information for getting the distance transmission. Next, the atmosphere light is estimated according to the distance transmission. Finally, the improved optical model is used to estimate the haze-free image. The experimental results show that our method is more effective and able to get better results than other single image based methods.

Segmentation of anatomical regions of a brain is the fundamental problem in pattern recognition in medical images. Lots of work has been reported, showing various level of accuracy in segmenting the boundary of an anatomy or tumor. The motivation for our work came from the fact of accurate delineation of the contour of a tumor from magnetic resonance images with high level of precision. We have developed an algorithm by modifying the existing Region Growing (RG) algorithm, by considering the local statistics of the pixels along with Pixel Run Length (PRL) parameter. PRL based Adaptive Region Growing (ARG) algorithm (PRL-ARG) gave satisfactory result with good level of accuracy. The segmented tumor is quantified by area, perimeter and form factor, which in turn helps to classify the different shape and contour of tumor. This algorithm is a semi - automated method and it will help the radiologist and neurologist to perform the diagnosis more effectively and accurately.

Almost all segmentation algorithms proposed in present papers have a common problem that the threshold value can't be accurately selected. A novel segmentation algorithm for fingerprint images is proposed in this paper, which is combining grey-value method and orientation diagram method with square-error method. When the difference between the grey-value of a ridge and that of a vale is great, foreground blocks of a fingerprint image are decided according to grey-level difference of more appearance frequency of pixels in histogram of the grey-scale. When the difference between the grey-value of a ridge and that of a vale is little, foreground blocks of a fingerprint image are decided according to ridges' orientation, which is that a grey-scale difference is calculated in each direction through 12-orientation template. The proposed method can segment well foreground blocks and background blocks of the fingerprint image. This method can segment foreground blocks and background blocks for the greater difference or less difference between fingerprint image ridge and vale accurately and easily. The boundary of a fingerprint image is obtained easily, while, the original image isn't destroyed. The experimental results show that the segmentation error rate of the proposed method is obviously lower than the square-error method and the improved method based on the square-error method in this paper.