Regional growth visual person color image segmentation
Source: Application of Electronic Technology: Zhang Wen Hao Chongyang Abstract: This paper tissues and organs color image segmentation algorithm based on region growing to extract the true color anatomical focus on the Visible Human data slice. By analyzing the color image color deck spatial RGB and HSV, color components suitable for pulmonary organ characteristics in visual person data are selected as feature metric as a growth algorithm. Experiments show that the method can better divide lung cancer organs in the visible people's true color chest image.
Keywords: regional growth visible person color image segmentation
Body visualization is a three-dimensional research area of information technology and medical disciplines in the late 20th century. Comprehensive development of the world's forefront of the salt industry is to integrate thousands of human body section data information in computer, redevelopment into a three-dimensional Stereoscopic image, an experimental platform that constitutes human morphological information research, providing a real model in providing a different type of human body related research, and has a major social application value. Visible Human Project is taken first in 1989 in 1989, 1994 and 1995, launched the world's first male and one female two frozen anatomical bodies high-precision, high-resolution real color organization Slice optical photos are accompanied by corresponding CT and MRI images as auxiliary data set. The Chinese visible human body was plans to officially launched in 2001. The 3rd Military Medical University visual human plan research project has launched China's first male and female corpse data set in 2002 and 2003. This data set has the advantage of smaller slice spacing, higher resolution, and greater capacity in the United States. At present, the human visualization has entered the stage of reconstruction and understanding of human body and organizational organs, in which the successful division of tissue organs of true color images is the core link that the project can be smooth, and it is also the most challenging problem.
The segmentation method of the color image mainly includes a color-specific cluster, Edge-based, regional generation, and the like. Among them, the colorful deck clustering is a multi-dimensional expand based on grayscale threshold, typically uses a threshold segmentation on each color component, and then combines the results. However, since the color histogram threshold is difficult to determine and the color mapping causes information dissipation, the cluster is usually overlapped with each other. The edge-based method is mainly to detect the boundary by using a mutation such as gradient or curvature line, but a complex medical image of the on the function, and the edge has a continuity of continuity. The method of regional generation is a type of method that is currently using more, effective effects; the disadvantage is that the selection of seed points will affect the results of the results, but for the complete use of automatic segmentation, it is still unpredictable medical anatomical image, and the artificial interaction selection The sub-point is a feasible method.
This paper selection 20 of the third military medical university visual person data set of women's bodies chest data, analyzed the color and texture characteristics of the pulse. The method of regional growth is taken, the initial seed point is selected by hand, and the pulmonary color characteristics is used as the segmentation criterion of the growth algorithm. The experimental results show that the method can better divide the lung region.
1 analysis
1.1 regional growth
The basic idea of regional growth is to set a pixel having similar properties to constitute a region. First, a seed point is selected as the starting point of the growth in each area to be divided, and then searches for those pixels that meet the similar feature of the seed point in the field of seed points, and combine with the region where the seed point is located. At this point, the newly merged pixels are used as the new seed point, and the above search and merge process are continued until the pixels can not be merged.
Let F represent a feature function that measures the degree of pixel, and the RK represents the area where the row is grown, the RK regional feature function is f (rk). Similarly, the current pixel is S (i, j), and its feature vector is f (s), the core of the regional growth algorithm is comparing the feature distance of F (RK) and F (S) satisfying a certain threshold T. If | f (rk) -f (s) | The biggest problem facing color image segmentation is to process five-dimensional information, two of which are geometric information, 3D is color information. A color distribution of a pixel can be represented by three-dimensional vector in color space. In the computer hardware, the color image is stored and displayed with RGB color space, and the change can be obtained from HSV, XYZ, YIQ and other color space. Among them, HSV is a user-oriented color model, closer to human eye to color, hue represents chromaticity, Saturation indicates saturation, Value represents strength. In the analysis of the lung data, the main feature of the lung difference between the lung difference in the real color image is found to be characterized in V = (R G B) / 3 component mean and variance, and the pulmonary The area has a large similarity of V components. Therefore, the intensity feature component of the pixel is thus used as the judgment method, and the RGB color space conversion formula is as follows: S = 1- {3 × [MIN (R, G, B)] / (R B G) V = (R B G) / 3 2 implementation 2. Reduce resolution A big challenge for processing visual person data is its large amount of data. Even in a small part of the body, we will maintain a complete resolution, and will take a lot of hard disk space and memory. For example, a true chest slice is 18MB, the resolution is 3072 × 2048, and the color depth is 24 bits. Therefore, the resolution must be reduced before segmentation. The author will drop to 1024 × 683 through every 3 × 3 pixels. Of course, the decrease in resolution can cause or enhance local effects. 2.2 Adaptive Region Growth Algorithm This article uses an adaptive regional growth algorithm. First, the operator selects a sub-point S (I, J) as the initial point in the region of interest (ROI), and then searches for 5 × 5 field φ, calculates the V component of this field. Means and variances, and the threshold THD and variance σ as judgment guidelines, the calculation formula is as follows: Average, where V is the V component value of the φ in the φ, the PV is the probability of component values equal to V. Threshold THD = RAVER-SV, where SV represents the component value of the V. The algorithm defines three operation sets, and the set CUR_SET is used to store pixel points in the divided area; the ready_set set the pixel point to be tested. Since the four adjacent pixel points are verified each time, the number of elements in the collection is always 4; Temporary collection TEMP_SET is used to store a new candidate center point. Initialization, only the initial seed point s, the four fields of S (I, J-1), S (I, J 1), S (I-1, J), S (I-1, J), S (I-1, J), S (I-1, J), S (I-1, J), S (I-1, J), S (I-1, J), S (I-1, J), S (I-1, J), S (I-1, J), S (I-1, J), S (I-1, J), S (I, J), S (I-1, J), S (I-1, J), S (I, J), S (I-1, J), i 1, J, Temp_set collection is empty. The program first searches for points in the READY_SET collection. If the difference between the component difference with the seed point is less than the threshold and the variance is less than 0.02, the point is placed in the CUR_SET collection, and the point Four adjacent pixel points are placed in a collection temp_set, waiting for the next step; otherwise, this point will be removed from the Ready_SET. Calculate until the Ready_set collection is empty, then the elements of the temporary collection TEMP_SET are seed points, establishing The new four-field point, wants a new Ready_set collection, the program is repeated above until the Teady_set collection is empty. 2.3 In the experimental results experiment, select the chest direct color image. Due to the large amount of data, the division of the entire chest must be done on the workstation. In order to improve the speed of operation and can be run on the PC, segmentation in the left and right lungs. At MATLAB 6.5, it is used 4 minutes and 56 seconds and 3 minutes to 40 seconds. The experimental results are shown in Figures 1 (a) to (d). As can be seen from Fig. 1, the algorithm is better to divide the pulmonary region, but the smoothness of the boundary needs further processing. In this paper, a preliminary study of the regional growth algorithm based on manually selecting seed points is used herein. The experimental results show that the method can better meet the accuracy requirements for the segmentation of the visible human lung region. However, there are some problems that need further exploration and resolution, including the extraction of pulmonary blood vessels, how to improve image resolution and accuracy while improving the operation speed, how to effectively promote algorithms to other organ parts.