Single Image Dehazing for Efficient Search Exploration Using Machine Learning Technique

Image dehazing is frequently used as a preprocessing step to enhance image clarity in various computer vision tasks, such as object tracking, intelligent surveillance and face recognition. It helps in the restoration of hazy images. Deep learning techniques have recently been utilized to significantly improve the quality of visual representation of images, but it takes a lengthy time to compute. In order to avoid the phenomenon of latency in the primary image analysis tasks, we have to monitor the time required for processing. An end-to-end modeling for real-time image dehazing network is proposed in this research paper. A complete network processing model for dehazing in real-time is proposed in this research paper. To record dependencies with reference to the positions and channels of the feature map, a non-local slice module is created. A region proposal network is also used to provide candidates with great sensitivity, hence raising true positive rates. The dark channel is estimated from the given input image along with the saliency of the detected image. The multi-scale feature maps are then used to suggest a false positive reduction module. An extensive comparative experiments were conducted using SANet's performance with a number of cutting-edge CNN-based detection techniques. The usefulness of our suggested method to gradually recover haze images utilising a coarse-to-fine strategy is demonstrated by evaluation results on SANet. The proposed method demonstrated real-time performance comparable to existing method in experimental findings using a publicly available dataset.