Novel no-reference multi-dimensional perceptual similarity metric.

Area of Science:

• Image processing and computer vision
• Medical imaging analysis
• Perceptual quality assessment

Background:

• Current image quality metrics, including deep learning and SSIM, require a reference image.
• This limitation hinders adaptive image acquisition, particularly in medical fields.
• A no-reference metric is needed to guide parameter adjustments during image capture.

Purpose of the Study:

• To develop a novel multi-dimensional no-reference perceptual similarity metric.
• To enable image quality assessment without a pristine reference image.
• To support adaptive image acquisition processes.

Main Methods:

• Combined a no-reference image quality metric (PIQUE) with perceptual similarity.
• Developed a multi-dimensional metric assessing noise, blur, and contrast.
• Validated the metric's performance through experimental correlation.

Main Results:

• The proposed metric effectively evaluates image quality without a reference.
• Demonstrated strong correlation between the metric and multi-dimensional image quality.
• Showcased the metric's utility in assessing noise, blur, and contrast variations.

Conclusions:

• The novel no-reference perceptual similarity metric offers a viable solution for quality assessment.
• This metric is beneficial for adaptive imaging, especially in medical applications.
• Future work can expand the dimensions of quality assessed.