Image Processing tutor : Martin Schmauder Marius PflĂźger Fraunhofer IPA
stud. Anca Mocan Politehnica University of TimiĹ&#x;oara Faculty of Electronics and Telecommunications
I. CATALYTIC CONVERTORS Using image processing in Matlab, we measure the amount of silicon carbide (SiC) in a catalytic converter, in order to find out if it can be recycled or not. In the following starting plot, we can notice a nonlinearity error at 0% SiC in the case of bright colored alloy convertors (green). Further on, I will try to suppress this error through 2 different algorithms.
Starting Plot
Algorithm I 1. Depending on the contrast limits: Considering that 100% SiC has a higher contrast image than lower values, the contrast means sharp changes between pixels of dark values and pixels of bright values. In the following plots, the green line shows the number of differences between low and high contrast limits vs. the amount of SiC in each converter.
Set 1
Set 3
Set 2
Set 4
Algorithm II 2. Eliminating the ‘noise’: In this case, after we convert the original image to an binary image using a threshold, we count the ‘white’ objects ( meaning no holes are considering ). In case of bright image, white objects are to many and we must eliminate the small ones (noise). This way we obtained a linear plot.
II. TISSUE ANALYSIS
Living Cells Detection An object can be easily detected in an image if the object has sufficient contrast from the background. We use edge detection, basic morphology tools and formulas to detect the living cells. We encountered some problems: 1. When closing the (white) objects, the close cells merge and we lose important data. 2. A histogram equalization resulted in a very bright, unclear image, rather than enhancing the objects. 3. Some cells were not clearly classified. 4. Dead cells have white outline that is also seen as an object.
Image Segmentation Algorithm: Steps 1. Read image and make dark pixels become ‘0’. 2. Edge enhancement using Laplace filter and dilate the resulted image 3. Complement and clean small objects, in order to extract only living cells. 4. Remove unimportant ‘noise’ and fill the object holes 5. Detect round objects using a ‘metric’ constant higher than 0.7 ( but around 1 ): metric = 4*pi*area/perimeter^2; 6. Count the round white objects Obs.! This algorithm eliminates noncells, dead cells or differentiated cells, but does not guarantee to count all of the living cells.
Edge detection & Dilation
Complement
Image clean & Object fill
Image clean & Object fill
References http://homepages.inf.ed.ac.uk/rbf/HIPR2/stretch.htm http://www.instructables.com/id/Image-Processing-and-Counting-using-MATLAB/?ALLSTEPS http://blogs.mathworks.com/steve/2006/06/02/cell-segmentation/ http://www.mathworks.de/products/demos/image/watershed/ipexwatershed.html#5 http://www.rpgroup.caltech.edu/courses/PBoC%20ASTAR/files_2011/Matlab/Image%20Analysis% 20with%20Matlab.pdf http://www.mathworks.de/de/help/images/ref/edge.html http://www.mathworks.de/products/image/examples.html? file=/products/demos/shipping/images/ipexcell.html http://www.rpgroup.caltech. edu/courses/PBL/bootcamp2011/protocols_and_references/AdvancedMatlabTutorial.pdf
The end... Danke schรถn!