# Introduction mage segmentation plays a crucial role in many medical imaging applications by automating or facilitating the delineation of anatomical structures and other regions of interest. Many methods for image segmentation proposed by researchers have some advantages and disadvantages related to its application and purpose [1]- [3]. Almost all of these methods focus on development of a method to improve image understanding by computer and/or develop image representation for extract computerized parameters of an image. But we must note that the best intelligent and complicated image processing machine is human vision system. With no doubt, all what we know and use today as image processing techniques is only a little projection of our vision system [4]- [5]. We in our research, first study nearly all of the current methods that are in use for image segmentation, then we with study the human vision system parameters in order to detect and extract segments of an image, propose the dominant parameters that are used by human vision system for segmentation and understand the image [3]. Finally we applied these parameters to detect the segments of a medical diagnosis mammogram database and detect the tumors in the mammograms. Author ? : Dept. of Electrical Engineering, Ecole Polytechnique, Montreal, QC, H3T1J4 Canada. E-mail : ekamrany@yahoo.com In this paper at first we survey current image segmentation methods, and mammogram image processing techniques. Then we based on Kandel theorem and Minima rule, implement some experiments to detect dominant parameters of human vision system in image segmentation. Then we introduce NCM points and textures detected using PCNN and LAW operators as dominant features in an image and try to extract them. After that we augment our detected and proposed parameters with NCM detection technique to implement our new algorithm. Finally we have developed our method on a standard database of mammography images and depicted the results. # II. # Mammogram image processing Breast cancer is the most common cancer in women. Early detection of the cancer leads to significant improvements in conservative treatment. We based on study the almost all current methods in mammogram image analysis saw that nearly all of these methods are focused on internal features of image rather than external features. # III. # Detecting NCM points Many objects have component parts, and these parts often differ in their visual salience. Based on the Kandel theory [6] which introduces edges as dominant features of an image, we have developed two different psychological experiments. In one of the experiments we understand the most effective factor of image edge that affect on human vision, and in the other one we found that the negative curvature minima (NCM) points are most effective points in an image that excite the human eyes [1]. IV. # PCNN and Texture Analysis We introduce the external feature on an image as dominant features and tried to detect them, but this is not enough to refuse the external features. So we select the PCNN and texture analysis using LAWs operators [7]-[9] as dominant external features that are much similar to and based on human vision technique for image segmentation. V. # Proposed method Detecting and extracting the dominant feature used by human vision system to understand segments June of an image and introducing the most similar image segmentation techniques to human vision were our purpose. We have combined the internal and external dominant detected features and introduced our new method and schema based on it. After preprocessing of image and extract it's contours, at firs we using cubic Bspline technique to fit a curve on image contours, then we find the NCM points and connect them using Euler spiral. Then we select the regions based on some introduced parameters such as Proximity, Co-circularity, Transparency, and Sharpness parameters. Finally we apply the PCNN and LAWs operators to the extracted region to detect the segments (tumors) more accurately. This technique also improves the efficiency of PCNN and LAWs operators by limiting the processing region to a small region. VI. # Implementation and Results In order to study and analysis the efficiency of our method, we used 200 mammograms of DDSM Data base. We designed a package called HMAM for implement our method. With introducing two parameters of TPR (True Positive Region) and FPR (False Positive Region) we measure these parameters in different images and compare the results with traditional methods. Some of the results are shown in Table 1 and Table 2. Some of the results of the applying proposed method on a cancerous mammogram are shown in figure 1. We repeat applying the method on images and saw that the results are dependent on number of iteration (figure 3). 2012![Global Journals Inc. (US) Global Journal of Computer Science and Technology Volume XII Issue X Version I](image-2.png "I © 2012") 123![Figure 1: Mass detection and extraction process regarding to the small (a) and big (b) masses](image-3.png "Figure 1 :Figure 2 :Figure 3 :") ![](image-4.png "FPR%") 1Star massesRegulatedmasses% FPR% TPR% FPR% TPRSmall ExtractsLow lev. Up lev11 580 B 19 881 B 5 1110 85Big ExtractsLow lev. Up lev1 1997 890 584 93 2size of extraction for the proposed methodStar massesRegulated Masses% FPR% TPR% FPR% TPRSmallLow Lev2 B 8233 B 014ExtractsUp lev.4 B 50885 B 1289BigLow lev.1021234ExtractsUp lev.1507215593 * Identification of Dominant Features in Image Segments Detection by Human Vision EKamrani MEnshaeyeh GHRad Proceedings of the 6th annual Iranian Computer Conference (CFSC2003) the 6th annual Iranian Computer Conference (CFSC2003)Tehran, IRAN Nov. 26-27, 2003 * Identification of dominant features of human vision for Image segments detection EKamrani MEnshaeyeh GHRad February 2004 proceeding to 9 th annual conference of computer association of Iran, Sharif university, iran * Parsing silhouettes: The short-cut rule MSGregory DSeyranian DHoffman Department of Cognitive Sciences University of California Irvine, California 1999 * Part boundaries alter the perception of transparency MSingh DHoffman 1998 California Irvine, California Department of Cognitive Sciences University of * Principles of Human NeuroScience ERKandel Jan 5, 2000 McGraw-Hill, Health Professions Division, USA * Neural Mechanisms of Scan Segmentation: Recordings from the Visual Cortex Suggest Basic Circuits for LinkingField Model REckhorn IEEE Transaction on Neural Networks 10 3 May 1999 * PCNN Models and Application JLJohnson MPadgett IEEE Transaction on Neural Networks 10 3 1999 * The use of texture analysis to delineate suspicious masses in mammography RGuptat P EUndrill February 1995 Forester hill Aberdeen AB9 ZZD, UK Department of Bio-Medical Physics and Bioengineering, University of Aberdeen