Integrating Local CNN and Global CNN for Script Identification in Natural Scene Images

Abstract

Script identification in natural scene images is a key pre-step for text recognition and is also an indispensable condition for automatic text understanding systems that are designed for multilanguage environments. In this paper, we present a novel framework integrating Local CNN and Global CNN both of which are based on ResNet-20 for script identification. We first obtain a lot of patches and segmented images based on the aspect ratios of the images. Subsequently, these patches and segmented images are used as inputs to Local CNN and Global CNN for training, respectively. Finally, to get the final results, the Adaboost algorithm is used to combine the results of Local CNN and Global CNN for decisionlevel fusion. Benefiting from such a strategy, Local CNN fully exploits the local features of the image, effectively revealing subtle differences among the scripts that are difficult to distinguish such as English, Greek, and Russian. Moreover, Global CNN mines the global features of the image to improve the accuracy of script identification. The experimental results demonstrate that our approach has a good performance on four public datasets.

An Auxiliary Fault Identification Strategy of Flexible HVDC Grid Based on Convolutional Neural Network With Branch Structures

“…Therefore, this paper proposes the structure of BR-CNN for fault identification, as shown in Fig. 1, where C layer represents convolution layer, P layer represents pooling layer, and F layer represents fully connected layer [21]. It can be seen that BR-CNN can set multiple input branches, so different fault characteristic parameters can be used.…”

“…Each weight matrix is obtained by training and learning from a convolution kernel. CNN extracts features of input data through multiple convolution kernels, so as to realize deep mining of features [21]. Taking the αth convolution kernel as an example, the calculation process from input layer to convolution layer can be expressed as…”

“…Therefore, it is difficult to distinguish lightning disturbance from positive PTG fault by analytical calculations. Due to the deep-level feature extraction capability of BR-CNN [21], it provides an idea to solve such problems that contain rich transient information, which can be used as an auxiliary criterion of the primary protection.…”

“…CNN mainly extracts high-level features of input data through convolution kernel. Because the convolution kernel is usually a two-dimensional matrix, it is especially suitable for feature extraction of twodimensional input data, such as the pixel matrix of a picture [21]. Therefore, with reference to image recognition, the voltage and current data on the positive and negative lines are arranged into two-dimensional matrixes.…”

“…The comprehensive evaluation index of the BR-CNN model is composed of two parts, including the overall accuracy rate and misjudgment rate. The overall accuracy rate is defined as (21)…”