segmentation of color image using adaptive thresholding and masking with watershed algorithm
TRANSCRIPT
Md. Habibur Rahman
American International University Bangladesh
Presented By
Md. Habibur Rahman* and Md. Rafiqul Islam
Authors
ICIEV 13, 17-18 May, 2013, Dhaka, Bangladesh
Main idea of our paper
• The main idea is to propose a modified version of the watershed algorithm for image segmentation
• An adaptive masking and a thresholding mechanism over each color channel before combining the segmentation from each channel into the final one.
• We have compared it with FCM, RG and HKM with respect to PSNR , MSE, PSNRRGB and CQM in 10 different kinds of images.
2
Overview of Image Segmentation
Image Segmentation is the method of assigning a label to each pixel in an image
The goal of image segmentation is to cluster of pixels in the relevant regions
Fuzzy C-Means (FCM)
Partition a finite collection of pixels into a collection of "C" fuzzy clusters
Region Growing (RG)
Group of pixels with similar properties to form a region
3
Overview of Image Segmentation (cont.)
Hill Climbing with K-Means (HKM)
This method detects local maxima of clusters in the global three-dimensional color histogram of an image
It associates the pixels of an image with the detected local maxima
Watershed (WS)
This method comes from geography
It is that of a topographic relief which is flooded by water
Watershed lines being the divide lines of the domains of attraction of rain falling over the region
4
Modified Watershed Algorithm
It can quickly calculate the every region of the watershed segmentation
Image normalization has been done by Eq. 1
5
Modified Watershed Algorithm (cont.) To determine the adaptive threshold by Eq. 2 and Eq.
3 based on Gray-threshold function
N-dimensional convolution for smoothing image
Adaptive masking operations by Eq. 4 and Eq. 5
6
Modified Watershed Algorithm (cont.)
Impose Minima to create morphological process image using Nucleus-masking (M2) on three color channels
Apply Watershed algorithm (Wn) on three color channels
Pixel labeling calculated by Ln = BWLABEL (Wn)
7
Modified Watershed Algorithm (cont.) Convert three channels into a RGB image for
visualizing the labeled regions by Pn = label2rgb (Ln)
R, G and B color channels (Pn) are added to generate segmented image
8
Overview of evaluation metrics
Peak Signal to Noise Ratio (PSNR) is calculated between two images by Eq. 6.
Mean Square Error (MSE) is calculated pixel-by-pixel by adding up the squared difference of all the pixels and dividing by the total pixel count using the Eq. 7.
Image Quality Measure (CQM) based on color transformation from RGB to YUV.
Reversible YUV Color Transformation (RCT) that is created from the JPEG2000 standard in Eq. 8
9
Overview of evaluation metrics (cont.)
The image quality metrics like PSNR of each YUV color channel (Y, U and V) is calculated separately
Finally, CQM value is calculated using the Eq. 9
Where, weighted luminance quality measure and weighted color quality measure components
Cw and Rw means the weights on the human perception of these cone and rod sensors
Cw and Rw are 0.0551 and 0.9449 respectively
10
Result Analysis
11
12
13
Conclusion Our proposed MWS method ensures accuracy and
quality of the 10 different kinds of color images
Proposed modified watershed approach can enhance the image segmentation performance
It is worth noticing that our proposed MWS approach is faster than many other segmentation algorithms, which makes it appropriate for real-time application
According to the visual and quantitative verification, the proposed algorithm is performing better than three other algorithms.
In future, we will focus on a more standard performance measure which could well reflect the difference between segmentation results 14
Questions ?
Thank You
for
Kind Attention
15