17 Image Dehazing Using Quaternion Complex Algebra-Based Neural Networks
N. Banupriya, G. Nirmala, N. Vijayaraghavan · 2024
Image dehazing is a technique used to reduce the effects of atmospheric haze in digital images. It is a difficult assignment because of the nonlinear nature of haze and the environmental conditions. To address this problem, we suggest a novel approach using quaternion complex algebra- based neural networks (QCANN). Unlike traditional CNNs, QCANN can efficiently deal with complex numbers, which might be better for representing the dynamics of haze in images. The proposed method first uses a haze density estimation algorithm to estimate the haze level in the input image. Then, the haze density map is transformed into a quaternion representation to include the complicated nature of the haze. This quaternion illustration is fed into the QCANN, containing parallel quaternion convolution layers and quaternion activation features. The QCANN is studied on a large dataset of hazy and haze-unfastened images to study the complex relationships between haze and image functions.