Learning a No-Reference Quality Assessment Model of Enhanced Images With Big Data

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Image quality assessment (IQA) and enhancement are crucial for applications like object detection.
• Raw images often require enhancement to improve visibility and contrast beyond their original quality.
• Machine learning approaches are increasingly vital in computational intelligence for image analysis.

Purpose of the Study:

• To develop a novel no-reference image quality assessment (NR-IQA) model.
• To establish a robust image enhancement framework guided by the developed NR-IQA measure.
• To improve visual quality of various image types through optimized enhancement.

Main Methods:

• A no-reference (NR) IQA model is proposed, extracting 17 features (contrast, sharpness, brightness) and using a regression module trained on big data.
• An image enhancement framework utilizes the NR-IQA measure to guide histogram modification for brightness and contrast correction.
• The model was validated against state-of-the-art methods on nine datasets.

Main Results:

• The proposed NR-IQA model demonstrates superiority and efficiency compared to full-reference, reduced-reference, and other NR-IQA methods.
• The enhancement framework successfully improves natural, low-contrast, low-light, and dehazed images.
• Experimental results validate the effectiveness of both the quality assessment and enhancement contributions.

Conclusions:

• The developed NR-IQA model offers a reliable and efficient method for assessing image visual quality without reference images.
• The quality optimization-based enhancement framework effectively rectifies image deficiencies, enhancing visual perception.
• This research contributes significantly to machine learning applications in image quality assessment and enhancement.