Handwritten Numeral Recognition Integrating Start–End Points Measure with Convolutional Neural Network

Abstract

Convolutional neural network (CNN) based methods have succeeded for handwritten numeral recognition (HNR) applications. However, CNN seems to misclassify similarly shaped numerals (i.e., the silhouette of the numerals that look the same). This paper presents an enhanced HNR system to improve the classification accuracy of the similarly shaped handwritten numerals incorporating the terminals points with CNN’s recognition, which can be utilized in various emerging applications related to language translation. In handwritten numerals, the terminal points (i.e., the start and end positions) are considered additional properties to discriminate between similarly shaped numerals. Start–End Writing Measure (SEWM) and its integration with CNN is the main contribution of this research. Traditionally, the classification outcome of a CNN-based system is considered according to the highest probability exposed for a particular numeral category. In the proposed system, along with such classification, its probability value (i.e., CNN’s confidence level) is also used as a regulating element. Parallel to CNN’s classification operation, SEWM measures the start–end points of the numeral image, suggesting the numeral category for which measured start–end points are found close to reference start–end points of the numeral class. Finally, the output label or system’s classification of the given numeral image is provided by comparing the confidence level with a predefined threshold value. SEWM-CNN is a suitable HNR method for Bengali and Devanagari numerals compared with other existing methods.

Improved EEG-based emotion recognition through information enhancement in connectivity feature map

“…After one or more convolutional-subsampling operations through a fully connected dense layer, the output layer categorizes the given 2D matrix as input of the CNN. The general description of CNN and its operations are available in existing studies, where CNN and its architectural issues are the primary concern 56 , 58 .…”

An algorithm for English capital letters based on discrete Hopfield network and BP network

“…According to the different selection of their activation functions, they are divided into discrete Hopfield networks and continuous Hopfield networks. In this paper, we use a discrete-type Hopfield network [10][11][12] (Discrete Hopfield Neural Network, DHNN). The network itself has the function of associative memory.…”