A Framework for image selection for image fusion using crowdsourced data

The proposed method in the paper actions the design an ensemble model based deep learning framework to choose the most relevant images from the crowd-sourced data for efficient image fusion. Image fusion is a technique of combining multiple registered images to get more informative or High Resolution (HR) image. Proposed method makes use of crowdsourcing to obtain multiple images. Crowdsourcing is a process of turning to a body of people to obtain relevant images. The images obtained through crowdsourcing may belong to different classes and may also contain multiple noisy and unwanted images and may be of different orientation. Also, the images obtained through crowdsourcing may be captured from different sources and also they may be in multiple resolution. The proposed method consists of training multiple deep learning models to estimate the probability of the image being relevant independently. Proposed method fuses the independent probabilities obtained from different algorithms using the decision fusion algorithm. Followed by elimination ofredundant images and the images containing significant amount of blur/noise. The selected images are then fused to generate the High Resolution (HR) image. The comparison of the results of proposed algorithms with different state-of-the-art fusion techniques using different qualitative and quantitative techniques yields increased efficiency through the proposed algorithm.