图像处理与模式识别.doc

Instructor: Ying Chen
Final project for the course of
“Image analysis and Pattern recognition”
钱平
信号与信号处理 s101904010
Project description
Through the figure 1 image analysis, we first need to adjust the brightness to enhance the fruit and background of contrast. Secondly, need to segment the different fruit images, generally use in the appropriate color space (such as HIS), from a certain color channel to segment the fruit, this method needs color space transformation and statistics a range of values, the calculation is relatively complicated. Image analysis technical classification of three basic category. Low-level treatment: image acquisition and pretreatment, do not need to intelligence. Intermediate treatment: image segmentation, representation and description, need intelligence. The advanced treatment: image recognition, explanation, lack of theory, in order to reduce the difficulty, design more special.

Fig.1

1 Principle
Fruit and background of edge represents the fruit of contour information, efficient, quick advantages has been widely used. We need to figure out the orange in different fruits. Through the analysis, and then RGB components have try out corresponding RGB components of the threshold value, if detected image color in this threshold is evenly into the same color within. If not in this threshold is besmear black within.

Fig.2 Recognition flowchart
2Proposed approach
2.1 Brightness adjusting
Due to obtain the images in the external environment and equipment not sure, causing image brightness uneven, influence subsequent edge detection, therefore, it is necessary to adjust brightness image. Here, take automatic brightness adjusting method, this method in the picture the biggest 5% pixel luminance taken out, then linear amplifier, make its average brightness 150. Normally, this method can realize uneven images of brightness good processing effects. In fig.1 after adjusting for brightness effect fig.3 shows.

Fig.3 Brightness adjusted diagram
2.2 RGB analysis
RGB color mode (also translated as "red, green, blue," less with) is a kind of color industry standard, is passed on the red (R), green (G), blue (B) three color channels changes and their mutual stack to get all kinds of color, RGB namely represents red, green and blue three channels of color, this standard nearly includes human visual perception of all colors that can be used at present, it is one of the most extensive color system. RGB color mode using RGB model for each pixel image in the RGB component to allocate a 0 to 255 range intensity values. For example: pure red R value for 255, G, value is 0, B value is 0, Gray R, G, B three values equal (except for 0 and 255); White R, G, B for 255, Black R, G, B is 0. RGB image using only 3 kinds of color, can make them according to different mixing ratio, on the screen recreate 16777216 colours.
At present, the edge extraction methods are classified into gray image edge extraction and color image edge extraction, the former using object and background of gray extraction, so inevitable will lose edge image color features and reduce the extracting accuracy. Research shows that the perception of color in the border, and played a leading role in color edge detection and monochrome edge detection, compared to achieve better effect. Color image edge extraction method has two kinds: output fusion method and multi-dimensional gradient method. Here, we select multidimensional gradient method, and directly in RGB channel processing, the detection process as shown in figure 4 below.

Fig.4 RGB analysis flowchart
2.3 Thresholds determined
The main principle is through pictures of RGB component comparison, then confirm corresponding reasonable threshold will find oranges. However threshold looking for is to more complex. I am from a picture a picture try out effect better threshold. Finally in threshold within the same kind of orange, part keep not threshold is within the grey value is 0.
3 Experimental results and analysis
First batch read image, then respectively with setting the threshold to compare. If the picture color in setting the scope remain the same kind of color, if not set range is all tu2 black. Finally the white or keep white. Due to the threshold is difficult to set up and given images have different brightness influence. Eventually lead to make the effect is not very good, a part of the oranges were also with a blackened, rather than oranges part but not been painted black.

Fig.5 The final result image
4 Conclusion
This paper is mainly using RGB color classification, and then setting threshold compared. Image points threshold in pixels if set threshold between, so these dot stays the same color, otherwise grey value be set to 0. An original is white still keep white. Due to the threshold set is not very accurate, yes processing after image effect is not very good. Orange part was also with a blackened, and need tu black part but not been painted black. This method, actually did not use directly through the training process, color features to extract the orange. Because no training process, will lead to identify the result is bad. Still need to modify and improve ideality effect, now only a rough identify oranges.
References
[1] He Dongjian , Geng Nan and Zhang Yikuan, in: Digital image processing [M]. Xi’An: Xian university of electronic science and technology press, 2008
[2] Li Guanzhang, Lu Qin etc, in: Color image brightness[J]. Computer engineering application,2010.
[3] R. C. Gonzalez and R. E. Woods, in: Digital image processing [M], Publishing house of electronics industry 2006.
[4] Liu Gang, in: MATLAB Digital image processing[M] . Mechanical industry press,2010.
4