Classification of Transformed and Geometrically Distorted Images using Convolutional Neural Network

Abstract

ConvolutionalNeural Network is a deep learning method that is used in many image-relatedapplications, such as image recognition and classification, it has achievedgreat performance in these fields, but it still suffers from some shortcomings.One of these shortcomings is not being able to be invariant to the input datadue to some image transformations like translation, rotation, scaling, andgeometric distortions such as skewness, perspective distortion and pincushiondistortion.  This study presents anoptimized CNN which uses the Geometric Heat Flow (GHF) to improve theperformance of the CNN regarding the invariant limitation and classificationaccuracy. GHF is a partial differential equation that expresses how the heatwould diffuse on a surface concerning time in a specific location. GHF isinvariant to image transformations and geometric distortions if it was takenconcerning the object's arc length which will lead to an invariant CNN. Theexperiments show that GHF improves the performance of the CNN, and the proposedwork achieves an accuracy of 98.09% on the MNIST handwritten dataset, 92.58% onthe MNIST-Fashion dataset, and 86.09% on the CIFAR-10 dataset.