Image Dehazing using Air Light Estimation and Modified IDeRs Model
Abstract - Images captured in hazy conditions, such as haze, fog, thin cloud, snow, silt, dust, off gas, etc., suffer from severe colour and contrast degradations. This includes remote sensing images (RSIs) as well as any other images captured in these conditions. As a result, there is a significant demand for dehazing algorithm to restore hazed RSIs from their degradations. The vast majority of dehazing algorithms that can be found in published works were initially developed for natural images dehazing (NID). In the context of our investigation, the physical model of NID is distinct from that of RSI dehazing (RSID), which has not been thoroughly discussed up until this point. In this paper, a novel idea referred to as "virtual depth" in relation to the physical model of RSI is presented for the first time. Real depth in a nature image is measured by the coverings of the earth's surface, such as snow, dust, cloud, and haze or fog. Virtual depth, on the other hand, gives the distance of an object departing from the foreground, whereas real depth measures the distance of an object departing from the foreground. These coverings serve the same purpose as the hazes in a natural photograph, namely to provide a hint of foreground and background. Dehazing operator is implemented iteratively to remove haze progressively until arriving at a result that is satisfactory in the second method, which is referred to as Iterative Dehazing for Remote Sensing image (IDeRS), which is proposed. In IDeRS, we also develop a fusion model for combining patch-wise and pixel-wise dehazing operators in order to eliminate halos and the oversaturation that is caused by them, respectively. This model is used to combine the two types of dehazing operators. The proposed IDeRS outperforms the majority of the state-of-the-art techniques in RSID, as shown by extensive experimental results tested on databases that are accessible to the general public.
Keywords - Dehazing, MATLAB, High resolution images, Natural image, IDERS, Air light Estimation.