Image dehazing network based on improved convolutional neural network

Image dehazing enhances its quality by restoring the actual pixels influenced by poor light and intensity due to environmental and other factors. Hazy images are rectified to improve visibility, guidance, and object recognition through channel attribute corrections. This article introduces a pre-emptive dehazing network (PDN) using an improved convolutional neural network (ICNN) for single to multi-image dehazing. In the proposed method, neural network layers are operated for intensity-based single and multi-feature analysis. The image is split based on intensity pixels for identifying the channel corrections. This channel correction and intensity verifications are processed using CNN in different independent layers. In the CNN training, the channel correction from the hidden layers and pixel correlation from the external dataset is performed for dehazing the image pixels. The dehazed pixels are organised based on the original input organisation for verifying the similarity measure. The proposed method's performance is validated utilising the metrics similarity, error, precision, F1-score, and time complexity.