Dear Shafaq

image it self is raw pixels 

first you have to extract some valuable information from image known as feature vector 

which you have to pass to SVM

Kindly provide me some samples or link if you can . I shall be very thankful to you

Dear Shafaq Nisar, 

Mere removing noise doesn't make your image (the matrix containing some numbers) suitable enough to be used as a dataset. The purpose of image pre-processing (noise removal etc) and image segmentation is to remove all the unwanted values from the matrix. But, image as such is a big jumble of numbers. The next task is to extract suitable information that may help the machine for recognizing the image. This is called 'Feature Extraction'. This is essential in order to get rid of 'the curse of dimensionality'. At this stage you get a new matrix containing less data and more information. This information can be fed in any classifier even to SVM, if it is a two class classification problem to recognize and extract the desired meaning of the image in plain text. 

The features will be varying according to you purpose and problem definition. For an OCR problem of printed english characters, your features might be size, shape, area, number of line crossings, std deviation, mean, median etc. among many others. If you are extracting 10 features for a particular image, your matrix that you will feed to the classifier will be 10x1 for that image instead of your original image which could have been a 564x564 matrix. Thus reducing a lot of data and extracting some meaningful information.

Now, regarding the choice of classifier, it is hugely dependent on the task that you want to accomplish.

Hope this helps. You may also study some basic OCR papers, tutorials on Image Processing and machine learning to build more understanding.