Outdoor images captured under adverse weather conditions such as haze, fog, and smog suffer from reduced contrast, color fidelity, and visibility, which hampers the performance of computer vision systems. This degradation is mainly caused by light scattering due to particles suspended in the atmosphere. Single-image dehazing methods have been widely studied to overcome this issue without requiring additional depth or multi-view data. This project implements a fast, efficient image dehazing algorithm based on the Dark Channel Prior (DCP) and morphological reconstruction techniques. The core contribution lies in replacing computationally intensive operations with morphological geodesic techniques that preserve essential image structures while accelerating the dehazing process. The pipeline is extended to include a text detection module using a convolutional neural network.
An object-wise evaluation is introduced to assess image quality for distinct regions, such as roads, trees, buildings, lights, and sky. Evaluation metrics like SSIM (Structural Similarity Index Measure), PSNR (Peak Signal-to-Noise Ratio), and execution time are used. The experimental results validate the effectiveness and efficiency of the proposed methodology compared to existing dehazing techniques.