Reverse Image Search Improved by Deep Learning

Reverse Image Search is also known as Instance Retrieval. This technology comes within applications of content based image retrieval. This technique enables developers and researchers to build scenarios beyond simple keyword search. A similar class of technology under the hood is used on sites to check original or similar photo for different purpose. Even face recognition in several security systems uses a similar concept to ascertain the identity of the person. For accurate and optimized similarity search this is done by feature extraction scaling the large image-databases. The image search operation should examine the contents of the image in featured vector by its pixel resolution ratio. One approach is to compare patches of areas between two images. In this paper Reverse Image Search by Google and many of the same techniques as the aforementioned are analyzed and improved by implementing the Deep Learning algorithms and image classifications. Deep learning method is to extract metadata from an image so that, it can then be scaled, indexed and used in a typical text query-based search. Cosine similarity is used to measure the angles of images in the high dimensional feature space. All test cases are implemented with the concept of deep learning using CNN (Convolutional Neural Network) in Python language with its tools Scikit-Learn, TensorFlow and OpenCV. This proposal has given with improved results.