Object Recognition in Images with Low-Resolution using Convolutional Neural Network

Abstract

Object recognition is a technology in computer vision that finds objects in an image or video series and identifies them. Phenomenal results have been recorded in object recognition studies using deep neural networks. But it has generally been deduced that sufficient image resolution and object size are obtainable, which cannot be assured in practical uses. Recognition of objects in lower resolution images is difficult. To overcome the stated problem, a Convolutional Neural Network (CNN) model for identifying objects in lower resolution images is proposed in this paper. In object recognition datasets, this approach outperforms the high recognition accuracy. In convolutional neural network models, both convolution and max-pooling layers are typically stacked. In the proposed approach, the pooling layer was substituted with a convolutional layer with an expanded phase without loss of precision in image recognition. The All Convolutional Neural Network with trained weights for recognizing lower resolution images is deployed. Through the obtained results, it is verified that the proposed model has high efficiency and accuracy.