Image Haze Removal via Reference Retrieval and Scene Prior

Area of Science:

• Computer Vision
• Image Processing

Background:

• Hazy scenes in photography exhibit low contrast, reducing visibility.
• Existing single-image dehazing methods have limitations due to reliance on specific priors or constraints.

Purpose of the Study:

• To develop an effective method for haze removal from single images.
• To leverage external information from correlated haze-free images for improved dehazing.

Main Methods:

• Utilized retrieved correlated haze-free images as external information.
• Performed global geometric registration and local block matching for spatial correlation.
• Combined external information with internal constraints and regularization to estimate the scene transmission map.

Main Results:

• The proposed method effectively removes haze, enhancing image visibility.
• Experimental results show superior visual quality compared to state-of-the-art methods.
• The approach successfully integrates scene structure and high-frequency information from reference images.

Conclusions:

• The novel dehazing method offers improved performance by incorporating external image data.
• This technique addresses limitations of traditional single-image dehazing approaches.
• The method provides a robust solution for enhancing visual quality in hazy photographic scenes.