ABF de-hazing algorithm based on deep learning CNN for single I-Haze detection

•Haze is a kind of natural event which is formed by smoke, dust, or other dry particles.•Therefore, the importance of removing the haze from the images is increased in the field of computer graphics or vision.•An Adaptive Bilateral Filter (ABF) is designed using an optimal selection of spatial weight parameters for de-hazing the scene depth of the hazy image.•Henceforth for detection purposes, a deep Convolution Neural Network is proposed to categorize the haze and non-haze images.•Here deep CNN is proposed for recovering the depth information in an input image using a network training and testing model. Haze is a kind of natural event which is formed by smoke, dust, or other dry particles. It acts as an overshot-on scene and degrades the visibility of environments and damages the image quality of indoor pictures due to the haze changing colors. Therefore, the importance of removing the haze from the images is increased in the field of computer graphics or vision. Because of its obscurity in mathematical formation, the removal process of haze becomes very complex and it is also made more difficult when the image is in a form of a purely single image. Here, the single-image haze removal process is the most complicated operation because of its ill-posed behavior. In this paper, the authors proposed a powerful single-image haze removal and recognition process. An Adaptive Bilateral Filter (ABF) is designed using an optimal selection of spatial weight parameters for de-hazing the scene depth of the hazy image. It is essential to classify both the haze and non-haze images because, without performing detection tasks, the non-haze images are also fed into the de-hazing process and lead to system complexity. Henceforth for detection purposes, a deep Convolution Neural Network is proposed to categorize the haze and non-haze images. Here deep CNN is proposed for recovering the depth information in an input image using a network training and testing model. The simulation results are demonstrating that the presented technique performed very well in the identification of the de-hazing effect and efficiency of haze removal algorithms.