Secure Content-Based Image Retrieval with Image Ambiguation

With the development of mobile devices, images and photos have become essential aspects of life for the majority of people. The storage space provided by these mobile devices frequently runs out due to the huge storage space requirements for the storage of these photographs. Because of this, the majority of users keep these data in the cloud. Additionally, the majority of service providers don’t use user-side encryption or ambiguation to protect the data. There needs to be another way to reduce the encryption overhead and reliably store the data without encryption and at the same time, preserve the image content. We use Image Ambiguation for this purpose. It aims at ambiguating the images with the help of a deep learning technique to distort the images in a way it can be securely stored on the cloud without any encryption. This technique reduces the encryption overhead and also reduces the chances of cloud providers gaining access to the stored data. We also need a reliable retrieval system. We propose an approach to solving the above problem by using bottlenecked autoencoders to reconstruct the images in such a way that the image on the cloud is unusable by the cloud provider and the storage used to store such images is kept to a minimum. To find the best matches for the query image, we use a retrieval technique based on VGG16 feature extraction and Euclidean distance with F1 scores of 0.902 and 1.0 on the COREL and CBIR Datasets respectively.