Now I understand your question.

1) create 6 matrices with the same size as your image. each matrix corresponds to each class.

2) for each pixel, suppose its probability for Class 1 is x and x in [0,1]. Set the element in Matrix 1, which corresponds to the same position as the pixel, to x*255.

3) save Matrix 1 as a gray-scale image

4) repeat 2) and 3) for all 6 matrices (classes). Finally, we obtain 6 gray-scale images, each of which is a probability map for each class.

I hope you will find this info helpful.

If i want to make a probability map of the image then how I can do that. Can someone help me on this topic

I believe no people can understand your question.

It seems your goal is to classify each pixel into 1 out of 6 classes. Questions: 1) is the image a gray-scale or a color image? 2) give some examples of the 6 classes 3) can you collect sufficient labeled pixels or images that can be used for training the classifier?

Possible solution:

For each class, collect known-class pixels labeled to this class, and train a GMM for this class by using the known-class pixels. Finally, we have 6 GMMs.

Given an unknown-class pixel, send it to the 6 GMMs and obtain 6 probabilities. Choose the one with the maximum probability.

I am working on face segmentation. I ave to classify six parts of the face(nose,mouth,hair etc), which i already did. As a result i have an image which has only six color for all the six parts. when i pass each pixel to a classifier i get six probability values for all six classes. Now my prof told me to make a probability map which will be a grayscale image. I hope now my question is clear.

thanks

Now I understand your question.

1) create 6 matrices with the same size as your image. each matrix corresponds to each class.

2) for each pixel, suppose its probability for Class 1 is x and x in [0,1]. Set the element in Matrix 1, which corresponds to the same position as the pixel, to x*255.

3) save Matrix 1 as a gray-scale image

4) repeat 2) and 3) for all 6 matrices (classes). Finally, we obtain 6 gray-scale images, each of which is a probability map for each class.

I hope you will find this info helpful.

BTW, classifying a "pixel" into a part of face sounds crazy to me. You'd better survey more recent papers to see if I am wrong. A better solution can be:

1) divide the image into small blocks, each of which has a size of say 8 by 8.

2) extract a HOG feature or a 128-dimensional SIFT feature from each block.

3) classify each block into a face part.

You really need to survey papers before doing any experiment.

I did the same Wu....I divided face images into small patches of 8x8 and then extracted three type of features from it. I passed the feature vector to classifier and then trained a classifier. Then i tested my images with a window of 8x8 but step size of 1. as a result i classify each pixel and got the images. I did all this. Now problem is just showing some probability maps of the images and that is what abt i am searching.

thanks for the nice explanation. I will try the method u told and will let you knw abt the result. Let me tell you some detail to clear the concept to you how i am doing my work.

1. I have two type of images, ground truth and original training images.

2. i divided training images into patches of 16x16 block with a step size of 8.

3. i extracted HSV,position and hog features from each patch and then passed each feature vector to classifier.

4. label was put on the training images using ground truth images.

5. similar process was did for testing i got realy good results.

now we have just to put some probability maps of some images in my paper to be published and for that i neeeded these probability maps.

Anyhow thanks alot for your cooperation and i will let u knw when i got the maps.

Xiaomeng Wu it worked. I got all the imges i needed. Thanks alot for your help. You solve a big problem of me

Xiaomeng Wu can we get probability map for an rgb image as well?

Hi Khalil,

What do you mean by "get probability map for an rgb image as well"?

Possibility 1: You did face-part classification using a gray-scale image, and now you want to do the same experiment using a color image.

Actually, I am not very familiar with face-part classification. Normally people just convert a color image into a gray-scale image, and then do classification. If you want to include color information into the feature, hmm..., I suggest you find the answer from related papers published in this field.

Possibility 2: You want to create a colorful probability map.

If you are using Matlab, suppose that the gray-scale probability map is M. A pseudo-code can be like:

colormap('jet'); imagesc(M);

If you are using OpenCV, check this link:

http://docs.opencv.org/trunk/modules/contrib/doc/facerec/colormaps.html

Please google the corresponding function names for more information.

BTW, I am happy that my comments can be helpful for you.

Hi Xiaomeng Wu,

Thanks alot. Your comments were really helpful for me and I am really appreciate that. I am working on face segmentation of color images using concept of semantic segmentation. Thank you so much Wu for your comments so quickly.