Hi friends,
Now I am working on NCR. I have used hog,phog,co-occurence hog for natural character recognition. Even i have tried deep learning with RBM. But I got 50% accuracy only so please suggest me a best feature extraction method.
In the attached paper a feature called Oriented Basic image Feature is compared with other features ( SIFT, ABBYY, Shape Context, HOG...). It may help you.
Conference Paper Natural Image Character Recognition Using Oriented Basic Ima...
Help us to help you and post some sample Images to give an Impression which Images you are dealing with.
I think your approach of using HoG is very appropriate, because HoG has always proven to be the best for natural scenes. However, I think you should consider combining HoG with Bag-of-word Modelling and a dimension reduction approach. Please, have a look at this paper, where we used HoG in combination with PLSA, and achieved good accuracies.
https://www.researchgate.net/publication/268812008_Unsupervised_Image_Classification_by_Probabilistic_Latent_Semantic_Analysis_for_the_Annotation_of_Images
Conference Paper Unsupervised Image Classification by Probabilistic Latent Se...
Thank you Abass. But i am not classifiying the image type, actually i am trying to recongize the characters. Below i have shared sample image.
https://www.researchgate.net/publication/257936623_Devnagari_Handwritten_Numeral_Recognition_using_Geometric_Features_and_Statistical_combination_Classifier
https://www.researchgate.net/publication/256846561_Development_of_Comprehensive_Devnagari_Numeral_and_Character_Database_for_Offline_Handwritten_Character_Recognition
https://www.researchgate.net/publication/49592833_A_Review_of_Research_on_Devnagari_Character_Recognition
Article Devnagari Handwritten Numeral Recognition using Geometric Fe...
Article Development of Comprehensive Devnagari Numeral and Character...
Article A Review of Research on Devnagari Character Recognition
Dear Nathan,
The state of the art natural image recognition algorithms are generally a variant of Deep Convolutional Neural Networks (CNN). Recently, these algorithms win Large Scale Visual Recognition Challenge 2014. Some of the algorithms reported as surpasses human-level performance. I write a summary and review on the following link:
https://fananymi.wordpress.com/2015/03/25/review-on-the-first-deep-learning-that-surpasses-human-level-performance/
https://fananymi.wordpress.com/2015/03/25/review-on-the-paper-which-reveals-the-power-of-convolutional-neural-net/
Hope this would help.
http://image-net.org/challenges/LSVRC/2014/results
- Badges
- Science topic