DL Based Genometric Image Retrieval

Retrieving images based on their characteristics has been a hot topic for decades. Deep learning (DL) methods are currently considered state-of-the-art for image processing, classification, and retrieval. Rotation is a frequent way to change the geometric shape of an image. However, due to the usage of anisotropic filters for convolution, Convolutional Neural Networks (CNNs) do not support rotation invariance characteristics. Therefore, the ability to rotate images without changing their essential qualities is a promising new area of research in the field of image processing. In this research, we suggest using feature-pretrained CNN model combinations for classification and comparable image retrieval; these models were trained on a large picture database that included images that had been rotated and scaled. This method yields excellent classification and retrieval results for the Corel dataset, both for the original query images and when rotated and scaled.