International Journal of Computer Engineering and Applications,

Transcription

1 COLOR IMAGE SEGMENTATION BY CLUSTERING APPROACH AND COUNTING THE NUMBER OF COLORS IN A COLOR IMAGE D. Jayasree 1, Ch. Rajasekhara rao 2, K. Krishnam raju 3 P.G. Student, Department of ECE, AITAM Engineering College, Tekkali, Andhra Pradesh, India 1 Associate Professor, Department of ECE, AITAM Engineering College, Tekkali, Andhra Pradesh, India 2 Assistant Professor, Department of ECE, AITAM Engineering College, Tekkali, Andhra Pradesh, India 2 ABSTRACT Image Segmentation plays a vital role in image processing. The research in this area is still relevant due to its wide applications. Image segmentation is a process of assigning a label to every pixel in an image such that pixels with same label share certain visual characteristics. This paper aims at counting the number of colors present in an image upon segmentation by clustering.. The image is segmented using a mean shift and normal cut approach and the number of colors are calculated. The results are compared among the two approaches. This type of approaches has applications in bio-medical image processing. KEYWORDS: Image segmentation, clustering, KDE, mean-shift, normal cut. [1] INTRODUCTION Color [1] is the quality of an object or substance with respect to light reflected by the object, usually determined visually by measurement of hue, saturation and brightness. Color vision [2] is the ability by humans to distinguish objects based on the wavelength or frequencies of light they reflect. The colors are perceived by the rods and cones present in human eye and can perceive more variations D. Jayasree, Ch. Rajasekhara Rao And K. Krishnam Raju 230

2 COLOR IMAGE SEGMENTATION BY CLUSTERING APPROACH AND COUNTING THE NUMBER OF COLORS IN A COLOR IMAGE in warmer colors than cooler ones. The number of colors a human eye can distinguish is about 10 million different colors. The human eye distinguishes and detects the colors by taking the information from the brain that process the image data. This led to the developments in computer vision systems to study the digital images characteristics. Digital image processing [3] is the subject that deals with the image analysis and image understanding which can be studied in the computer vision systems through certain software, where a digital image can be processed and properties related to it are studied. There are still many visual tasks humans can easily do, but that are beyond the reach of computer vision systems. There are many techniques in digital image processing to study about the image characteristics and it s color properties and one of the contemporary method is through image segmentation. Image segmentation [4] classifies or partitions an image into several parts (regions) according to the image, example, the pixel value or the frequency response. There are lots of image segmentation algorithms which are extensively applied in science and computer vision applications. We can categorize them as region-based segmentation [10], edge-base and data clustering segmentation [11] etc. Image segmentation algorithms are based on one of the two basic properties of the intensity value i.e. discontinuity and similarity. There are different approaches for different type of images. The first approach represents Histogram thresholding, second approach is Edge based and the last one is region based approach. In histogram thresholding different gray or color ranges are represented to made regions of an image. In the second approach, different edge detection operators are used and also the edges are joined if the regions are not connected. In the third approach images are partitioned into regions which are similar according to a set of predefined criteria [9] There are number of colors present in an image and a few of them can be countable manually. In this paper the numbers of colors present in a color image are extracted through clustering [5]. The segmentation process where the pixels of same features are grouped or clustered into one group is called clustering. If we acquire the number of clusters it gives the number of colors present in an image. The image is clustered by mean shift [6] and normal cut [7] approach and the numbers of clusters are calculated respectively. There are some methods to count the colors in the image directly which may not give the accurate number. This project approach is to segment the image into clusters and extract the numbers of colors present in that image which gives better result than other unsupervised techniques. Extracting the number of colors from the image has application in medical image analysis[8] to detect the cancerous and tumor cells. This paper is organized in four sections. Section I give an introduction. Steps for proposed method are discussed in section II. Section III discusses the results. Section IV gives Conclusion. D. Jayasree, Ch. Rajasekhara Rao And K. Krishnam Raju 231

3 [2] FLOW CHART OF PROPOSED METHOD As the image segmentation is an emerging field in digital image processing in some cases it becomes necessary to find the total number of colors from the original image. It is very useful in many medical applications some of which are detecting brain tumors and cancerous cells. In the proposed method image with total number of colors is presented. The procedure of calculating total number of colors from an image is very helpful in quantization of image. For k-means clustering first, an image is taken as an input. The input image is in the form of pixels and is transformed into a feature space (RBG). Next similar data points, i.e. the points which have similar color, are grouped together using any clustering method. A clustering method such as k-means clustering is used to form clusters as shown in the flow chart. The distances are calculated using and Euclidean distance. The data points with minimum distance or Euclidean distance are grouped together to form the clusters. After clustering is done, the mean of the clusters is taken. Then the mean color in each cluster is calculated to be remapped onto the image. The biggest disadvantage of our heavy usage of k-means clustering, is that it means we would have to think of a k each time, which really doesn t make too much sense because we would like to algorithm to solve this on his own. In the proposed method the flowchart depicts the step by step procedure. The steps are preprocessing, the input image is converted from RGB to gray scale. Then the image is filtered to remove any noises present in the image. Here mean shift algorithm is used in the image segmentation process where it decreases the complexity due to k-means, where the value for k has to be given. Now the image segmentation using mean shift clustering technique and total clusters present in the image are calculated automatically. Normal cut, the common theme underlying these approaches is the formation of a weighted graph, where each vertex corresponds to n image pixel or a region, and the weight of each edge connecting two pixels or two regions represents the likelihood that they belong to the same segment. The weights are usually related to the color and texture features, as well as the spatial characteristic of the corresponding pixels or regions. A graph is partitioned into multiple components that minimize some cost function of the vertices in the components and/or the boundaries between those components. The same procedure is applied again with different segmentation algorithm which is normal cut and the total number of clusters present in the image is calculated. This is done in order to compare the results within the two algorithms that are used for clustering. D. Jayasree, Ch. Rajasekhara Rao And K. Krishnam Raju 232

4 COLOR IMAGE SEGMENTATION BY CLUSTERING APPROACH AND COUNTING THE NUMBER OF COLORS IN A COLOR IMAGE Start Input image (choose from the directory) Pre-processing (converting to gray scale) Filterization KDE Image segmentation by mean shift algorithm Image segmentation by normal cut algorithm Output image (total no. of colors) Output image (total no. of colors) Exit Steps for meanshift clustering: 1. Read the input color image. 2. Create a vector X whose rows are the RGB values of the image pixels. 3. kde is the step that is required before applying mean shift algorithm. 4. Apply the mean shift algorithm to X with a required kernel width. 5. Assign to each pixel the value of its cluster center and obtain the segmented image. D. Jayasree, Ch. Rajasekhara Rao And K. Krishnam Raju 233

5 Steps for counting colors: 1. Read the input color image. 2. Convert image from rgb to gray. 3. Extract the planes with rgb colors and store them into variables. 4. Form a single matrix to see values of each pixels. 5. Total number of colors existing in an original image is calculated. [3] RESULTS AND DISCUSSIONS We applied our algorithm on various color images and as a result we get total number of colors from the given color image. The below figures depicts the original color image and the other images which are shown below contains the extraction of total number of colors from an original image upon image segmentation. The main feature of this project is that, it can calculate the colors in an image automatically which makes the process faster as compare to other methods. (a)original image (b)mean shift using color information D. Jayasree, Ch. Rajasekhara Rao And K. Krishnam Raju 234

6 COLOR IMAGE SEGMENTATION BY CLUSTERING APPROACH AND COUNTING THE NUMBER OF COLORS IN A COLOR IMAGE (c)meanshift using color and spatial information (d) normal cut using spatial informtion (a)original image (b) mean shift using color information (c) normal cut using color and spatial information (d) normal cut using spatial information Table for comparison between the number of colors obtained from the color image. Type of image K-means Mean-shift(using color information) Mean-shift(using color and spatial information) Lena Horseand pony Sea and rocks cheetah Normal-cut D. Jayasree, Ch. Rajasekhara Rao And K. Krishnam Raju 235

7 [4] CONCLUSION The proposed method counts the total number of colors through the image segmentation based on clustering. This method is applied to different images and the results are obtained. This technique is capable because we can get total number of colors from an image automatically which is amazing. If we know the total number of colors then it can be very helpful in medical image processing applications such as identifying the tumor cells and cancerous cells in the medical images. REFERENCES [1] A Review: Color Models in Image Processing Harmeet Kaur Kelda GNDU, Assistant Professor, Amritsar, India. [2] The machinery of colour vision Samuel G. Solomon & Peter Lennie. [3] Digital Image Processing: an Overview by E. Lyons [4] Image Segmentation Techniques Rajeshwar Dass, Dept. of ECE, DCR University of Sci.& Technology, Murthal, Sonepat, Haryana, India. [5] A Review on Image Segmentation with its Clustering Techniques Priyansh Sharma and Jenkin Suji, Dept. of EC, ITM University. [6] Image Segmentation based on Mean Shift Algorithm and Normalized Cuts C.Hari Hara Suthan, Dr.R.V.S.SatyaNarayana student, Dept. of ECE, SVU College of Engineering, Tirupati, Andhra Pradesh, India. [7] Normalized Cuts and Image Segmentation Naotoshi Seo, November 8, [8] Detecting and counting the number of white blood cells in a blood sample images by color based k-means clustering Neha, Sharma, IJEEE, Vol. 1, Issue 3 (June, 2014). [9] Navneet Kaur et.al Bacteria foraging based image segmentation An International journal of Engineering Sciences ISSN: Issue July 2012, Vol. 6. [10] Y. B. Chen and O.T.C. Chen, Image segmentation method using thresholds automatically determined from picture contents, EURASIP Journal on Image And Video Processing, Article ID140492, 2009, doi: /2009/ [11] Image Segmentation by Clustering Methods: Performance Analysis Volume 29 No.11 B. Sathya, Department of Applied Science Vivekanandha Institute of Engineering and Technology for Women Thiruchengode, Tamilnadu, India R. Manavalan Department of Computer Science (PG) K.S.R College of Arts and Science Thiruchengode Tamilnadu [12] A Study Of Image Segmentation Algorithms For Different Types Of Images, Krishna Kant Singh, Akansha Singh Dept. Of ECE, Hemes Engineering College Roorkee, India Dept. Of Information Technology, AKGEC Ghaziabad, India. D. Jayasree, Ch. Rajasekhara Rao And K. Krishnam Raju 236