Gradient Information-Orientated Colour-Line Priori Knowledge for Remote Sensing Images Dehazing

Abstract

The frequent occurrence of haze weather seriously affects the object detection and other remote sensing applications, so it is a great challenge to recover clear objects in haze image. Image dehazing can make the raw image have higher contrast, sharpness and more detail information. Therefore, it is of great significance to study image dehazing. In this paper, we propose a gradient information-orientated colour-line priori knowledge method for remote sensing images dehazing. In the case that the color-line prior is not applicable to some scenes, in this paper, a transmittance optimization method of edge-preserving is proposed. The weight distribution is obtained by calculating the prior confidence of the color-line. The similarity between the transmittance of different pixels and the prior image is controlled by the weight. Meanwhile, the gradient information of the original image is used to control the transmittance edge to avoid halo effect. This method can estimate the haze distribution more accurately, thus it can recover a clear image without haze. The experimental results show that our method obtains the better effect in terms of Structural Similarity Index (SSIM), Edge Preservation Index (EPI), mean squared error (MSE), Information Entropy (IE), Gray Mean Grads (GMG) than other state-of-the-arts image dehazing methods.