NIMA: Neural Image Assessment.

Area of Science:

• Computer Vision
• Machine Learning
• Image Processing

Background:

• Automated image quality assessment is crucial for various applications, including evaluating image capture, storage, and media sharing.
• Existing methods often predict only the mean opinion score, failing to capture the full spectrum of human perception.
• The subjective nature of image quality poses a significant challenge for automated systems.

Purpose of the Study:

• To develop a novel method for image quality assessment that predicts the distribution of human opinion scores.
• To create a simpler yet high-performing architecture for no-reference image quality evaluation.
• To enable semantic- and perceptually-aware quality assessment without requiring a reference image.

Main Methods:

• Utilized a convolutional neural network (CNN) architecture.
• Retrained state-of-the-art deep object recognition networks for the quality assessment task.
• Focused on predicting the distribution of human opinion scores rather than just the mean score.

Main Results:

• Achieved comparable performance to existing methods with a significantly simpler architecture.
• Demonstrated reliable image scoring with high correlation to human perception.
• Enabled no-reference quality assessment, adaptable for optimizing photographic enhancement algorithms.

Conclusions:

• The proposed CNN-based approach effectively predicts the distribution of human opinion scores for image quality.
• This method offers a simpler, more robust solution for no-reference image quality assessment.
• The developed network aids in optimizing image enhancement algorithms and provides perceptually aware evaluations.