2
$\begingroup$

I am working on OCR. I have extracted features of 26 alphabets, and I want to compare them with an experimental character feature.

I have tried using Euclidean distance and correlation functions, but I am not getting good efficiency. What are some alternate methods for the same?

$\endgroup$
1
  • 1
    $\begingroup$ When the euclidian distance fails, the next reasonable one is the Mahalanobis distance. It normalizes the euclidian distance in each dimension by the dispersion of the data (of the variable under test) for that dimension. In the most complete version of the Mahalanobis distance, the cross correlation among dimensions are also taken. The concept is simple, and the results might adequate for you. In addition, you might want to reduce the dimensionality of your data using PCA, or kernel-PCA. Please share your results, if any. $\endgroup$ Jan 2 '17 at 19:35
2
$\begingroup$

As far as I understood you're seeking the best similarity measuring function. There are zillions of metrics for that purpose, in fact any clustering algorithm such as SVM, K-means and neural network based approaches in one or other way do that. Take a look at these links :

https://reference.wolfram.com/language/guide/DistanceAndSimilarityMeasures.html

http://nebc.nerc.ac.uk/courses/GeneSpring/GS_Mar2006/Measures%20of%20Similarity.pdf

$\endgroup$
0
$\begingroup$

If you just need OCR to work, then please try the ocr function in the Computer Vision System Toolbox.

If you need to write your own, take a look at this example, showing how to train an SVM classifier using HOG features for recognizing hand-written digits.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.