Decoding Image Similarity: ResNet vs ViT

In today’s time, the ability to search for visually similar photos has become very important especially when digital images are created so fast. This paper tries to show an approach for similar image searching, mainly focusing on locations, architectural styles, and remote sensing data. We aimed to find similar features in images but without needing those big training datasets. By using advanced methods, such as ResNet and Visual Transformers. Resnet is great at capturing small details like textures and edges, while Visual transformers, help understand complex relationships in images with attention process. Together, they work well for detecting similarities. For evaluation, pre-trained models and similarity metrics were used, such as Normalised Cross Correlation (NCC) and the Structural Similarity Index (SSIM). This way, we could skip training models from scratch. The findings were checked using Mean Opinion Score (MOS), which proved this approach works. This method is especially useful in things like urban planning, historical research, and monitoring terrain changes with remote sensing. For example, it's helpful to find similar architectural buildings or notice changes in landscapes. Our study shows that this approach might improve how images are searched and used for making better decisions in these areas.