Please use this identifier to cite or link to this item:
|Title:||A novel classification of handwritten digits using compressive sensing technique|
|Abstract:||This paper proposes a novel method of feature extraction for English digit recognition using compressive sensing. Compressive sensing provides the advantage of capturing the image by taking few measurements. These few measurements are capable of perfectly reconstructing the original image. Perfect reconstruction occurs because those few measurements capture the essence of the image. Hence, we used those few measurements as the features for digit recognition. Finally, we train two classifiers namely Support vector machine (SVM) and k-nearest neighbor (k-NN) with the resulting features and obtained performance of 95.0535% and 94.9300% respectively for both the classifiers on MNIST database. The purpose here is not to achieve a better accuracy than the other existing methods but to show that compressively sensed samples can be used as a feature for multiclass classification with more than 95% accuracy. After that we performed the same task by adding noise to the training images to show the robustness of compressively sensed features. At last we concluded that capturing the full image and then extracting features is as good as compressively sensing the image and perform pattern recognition on that compressed measurement. The results are quite good but on a new paradigm. � 2016 IEEE.|
|Appears in Collections:||Research Publications|
Files in This Item:
There are no files associated with this item.
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.