Improved firefly algorithm-based optimized convolution neural network for scene character recognition

Abstract

The most common and challenging issues in image recognition are scene character recognition from the street view image, and the scene character consists of both text and number. Recently, the researchers were introduced a lot of scene character recognition methods, but the performance of the methods often degraded due to complexity. So, we proposed the improved firefly algorithm for local trapping problem (IFLT) utilizing convolutional neural network (CNN) for the extraction of features from the scene character. The IFLT approach is the improved version of the firefly optimization algorithm to solve local trapping problems. During feature extraction, the hyperparameters on CNN are tuned with the help of the IFLT approach. The alignment and multilayer perceptron layers are used on CNN. Subsequently, the support vector machine approach is used to classify the relevant class of scene characters from the street view image. Experimentally, we use six scene character dataset SVHN, ISN, IIIT5K-words, SVT, ICDAR 2003, and ICDAR 2013 dataset. The performance of the proposed IFLT approach is evaluated with standard deviation, mean, average computational time, and most excellent minimum (MEmin) parameters. The experimental results demonstrate that the proposed IFLT-CNN is well suitable for scene character recognition.