Haze in the air degrades actual imaging effect, directly leading to low contrast and clarity of images, making it difficult to reflect the real scene information, thereby reducing the application effectiveness of pattern recognition, computer vision, and intelligent detection. Therefore, image dehazing is necessary. To improve foggy images, scholars have proposed image enhancement-based, image restoration-based, and deep learning-based dehazing methods. However, image enhancement-based methods are subjective and arbitrary, lacking robustness across different foggy images. Image restoration-based methods model hazy imaging using assumptions or priors, which struggle to fit complex real-world scenarios accurately, leading to deviation between restored images and the actual scene images. Deep learning-based methods learn models from image datasets, but they require large datasets and long training time.

To address these shortcomings, an image dehazing method based on adaptive transmittance and Gamma enhancement was proposed. Firstly, an adaptive dark channel was defined: for pixels in sudden-change regions, the dark channel was calculated using a minimum filter with a smaller neighborhood, while for pixels in smooth regions, it was calculated using a minimum filter with a larger neighborhood. This adaptive dark channel exhibited no block effect, had delicate edges and textures, accurately reflected brightness variations, and enhanced image hierarchy. Secondly, based on the adaptive dark channel, adaptive transmittance was proposed. Pixels were classified into sudden-change and smooth regions based on local standard deviation, and their transmittance was calculated using minimum filters with small and large neighborhoods, respectively. Thirdly, the top 1 ‰ of pixels with the highest values in the dark channel were selected, and their corresponding pixel values in the foggy image were averaged within each channel to estimate atmospheric light, improving estimation accuracy. Finally, dehazing recovery was performed using the estimated adaptive transmittance and atmospheric light based on the atmospheric scattering model. Furthermore, to restore the normal brightness of dehazed images, an adaptive Gamma transformation function was proposed, applying the Gamma transformation to the brightness component in HSV space to improve brightness distribution. The defined Gamma exponential function, as shown in Equation (8), adapted to the grayscale frequency distribution of the brightness image. Pixels were assumed to be evenly distributed across gray levels, with the denominator being the integral of each gray level multiplied by its corresponding average frequency, and the numerator being the integral of each gray level multiplied by its corresponding pixel frequency.

The proposed method and several recently developed representative methods were applied to six scene-based image dehazing experiments, as shown in Fig.4 to Fig.9. Subjectively, the proposed method achieved more thorough dehazing, resulting in higher contrast, clarity, and better color fidelity, leading to superior visual effects. Objectively, the information entropy of images dehazed by the proposed method was over 4.01% higher than that of existing methods, demonstrating that the adaptive transmittance enhanced information content, yielding clearer textures and details. The standard deviation of dehazed images was over 23.1% higher, indicating greater improvement in contrast and clarity. The average brightness was over 6.67% higher, showing that the adaptive Gamma enhancement effectively improved brightness distribution without over-enhancement.

Experimental results and qualitative-quantitative analyses demonstrate that the proposed method achieves competitive dehazing performance, yielding high-contrast dehazed images with clear edges, rich details and realistic colors.Therefore, the proposed image dehazing method based on adaptive transmittance and Gamma enhancement is effective.