Robust Handwritten Character Recognition with Features Inspired by Visual Ventral Stream

Abstract

This paper focuses on the applicability of the features inspired by the visual ventral stream for handwritten character recognition. A set of scale and translation invariant C2 features are first extracted from all images in the dataset. Three standard classifiers kNN, ANN and SVM are then trained over a training set and then compared over a separate test set. In order to achieve higher recognition rate, a two stage classifier was designed with different preprocessing in the second stage. Experiments performed to validate the method on the wellknown MNIST database, standard Farsi digits and characters, exhibit high recognition rates and compete with some of the best existing approaches. Moreover an analysis is conducted to evaluate the robustness of this approach to orientation, scale and translation distortions.

Augmenting the generalized hough transform to enable the mining of petroglyphs

“…We obtained the best error rate of 4.54% in the resolution of 20×20 (the same as the best resolution for the NicIcon dataset). Borji et al [6] performed extensive empirical tests on this dataset, testing multiple algorithms, 3-NN, ANN, SVM polynomial , SVM linear and SVM RBF , each with four parameter choices (two choices of filters times two numbers of orientations). Of the twenty reported error rates, the mean was 8.69% and only four combinations beat our approach with a best performance of 2.36%.…”

A Shape Descriptor Based on Trainable COSFIRE Filters for the Recognition of Handwritten Digits

“…The recognition rate of 99.52% that we achieve for the MNIST data set is comparable to the best existing approaches 4 . In particular, our method outperforms the shape context approach (99.37% in [4]), and three other approaches (94.20% in [17], 97.62% in [5] and 98.73% in [10]) that use biologically inspired feature detectors combined with a multi-layer perceptron (MLP) [17] and a linear SVM classifier [5,10]. The highest recognition rate achieved to date is 99.73% [7].…”

“…For the Farsi data set, the recognition rate of 99.33% that we achieve is better than the best results ever reported in the literature. The COSFIRE filters outperform other biologically inspired feature detectors (96% in [5] and 99.1% in [10]) combined with a linear SVM classifier and also outperform the modified gradient technique which resulted in a recognition rate of 98.8% when combined with a multiple classifier system based on MLP classifiers [13].…”

“…A number of methods have been proposed, including Zernike moments [3], direct matching [9], Fourier descriptors [12], geometric moment invariants [21], shape context [4], zoning [8] and biologically motivated features [10]. For a thorough overview of recent advances in the recognition of Western Arabic and Farsi handwritten digits we refer the reader to [16,22] and [20,5], respectively.…”

An efficient and effective similarity measure to enable data mining of petroglyphs

“…Borji et al (2008) performed extensive empirical tests on this dataset, testing multiple algorithms, 3-NN, ANN, SVM polynomial , SVM linear and SVM R B F , each with four parameter choices (two choices of filters times two numbers of orientations). Of the twenty reported error rates, the mean was 8.69%, and only four combinations beat our approach, with a best performance of 2.36%.…”