Batch Normalized Siamese Network Deep Learning Based Image Similarity Estimation

Abstract

The assessment of how two distinct images are equal are indeed called image similarity and consistency. In other words, it measures how much the intensity patterns in two images are comparable to one another. In order to achieve this, researchers examine the image descriptors recursively in order to identify descriptor pairs that are comparable. The two images are deemed comparable if the number of related descriptors exceeds a predetermined threshold and both images exhibit the very same entity. The computation of image similarity is used for various applications which graves to be the mandatory process for production of the application. With this intent, the Fashion MNIST dataset from KAGGLE is used for implementing the image similarity estimation. This paper proposes Batch Normalized Siamese Network (BNSN) deep learning based model for computing the image similarity. The BNSN model is designed with two subnetworks that generates feature vectors of two input images. The lambda batch normalization is performed with single dense layer to predict the image similarity with label 0 indicating the identical images and label 1 denoting the different images. The 30,000 training images were fitted with BNSN and tested with 30,000 images. Python was implemented on a Geforce Tesla V100 NVidia Graphics card webserver with a batch size of 64 and 30 training epochs. The training images are also tested with traditional image similarity method and implementation of proposed BNSN shows the accuracy of 91.91%, Precision of 92.93%, Recall of 90.72% and FScore of 91.81%.