Image Quality Analysis: Full Reference

0
1941

The digitization of cultural heritage is an essential practice in the preservation and access to historical and artistic objects. However, ensuring that digital representations are faithful to the originals is a significant challenge. To address this challenge, image quality metrics based on full references are used, allowing an objective evaluation of the fidelity and quality of digitized images. This discussion will focus on the importance of these metrics in the analysis of image quality in the digitization of cultural heritage, with reference to ISO 19263-1 and ISO 19264-1 standards.

Introduction to Reference-Based Metrics

Full-Reference Metrics are used to compare the digitized image with a reference image considered “perfect” or free of distortions. These metrics are fundamental in processes where images are manipulated to assess the preservation of quality and content fidelity. Although not always feasible in all scenarios due to differences in media (image-matter and image-code), they are particularly useful in specific contexts such as video encoding and transmission.

Fidelity Metrics

Fidelity metrics estimate the pixel-level similarity between the reference image and the derived image. The two most common metrics in this category are the Mean Squared Error (MSE) and the Peak Signal-to-Noise Ratio (PSNR).

MSE (Mean Squared Error)

MSE calculates the mean of the squared error between corresponding pixels of two images. This value is always positive, and an MSE of zero indicates the total absence of error.

Although MSE provides a direct measure of error, it can result in values that are difficult to interpret due to its quadratic nature. To address this, the Root Mean Squared Error (RMSE) is often used. RMSE returns the error to a more comprehensible scale by taking the square root of the MSE.

PSNR (Peak Signal to Noise Ratio)

PSNR is expressed in decibels and measures the ratio between the maximum power of a signal and the power of the noise affecting its representation. It is widely used in the video sector, as well as in image quality studies due to its ability to correlate with subjective quality assessments.

Quality Metrics

Unlike fidelity metrics, quality metrics such as the Visible Differences Predictor (VDP), Video Quality Metric (VQM), and Structural Similarity (SSIM) consider the perception of the human visual system, providing a more comprehensive evaluation of how differences between images will affect human perception of quality.

VDP (Visible Differences Predictor)

VDP, developed by Scott Daly in 1992, evaluates the physically visible differences between pairs of images. This metric is crucial because metrics like MSE and PSNR do not always accurately reflect visible defects that the human eye can detect. For example, a small difference in MSE values might be invisible to the human eye, while a significant difference could be easily perceptible.

VQM (Video Quality Metric)

Developed by Margaret H. Pinson and Stephen Wolf, VQM has an excellent correlation with subjective video quality evaluations. It has been recognized as a standard (ANSI T1.801.031996/2003) and adopted by ITUT J.144. This metric evaluates video quality considering various perceptual features, making it especially useful in video transmission.

SSIM (Structural Similarity)

Proposed by Zhou Wang in 2004, the SSIM metric evaluates image quality based on the comparison of local structures. Unlike MSE and PSNR, which do not consider how pixels are perceived interdependently, SSIM measures alterations in terms of luminance, contrast, and structure. This allows SSIM not only to quantify the total difference between two images but also to generate difference maps that locate specific areas of discrepancy between images. This is particularly useful for detecting compression artifacts and other defects.

Conclusion

Full-reference metrics are crucial for evaluating the quality and fidelity of digitized images of cultural heritage. These metrics provide objective tools to measure differences between the digitized image and an ideal reference. Standards such as ISO 19263-1 and ISO 19264-1 provide a robust framework for consistent and high-quality digitization, ensuring that images maintain the necessary integrity for long-term preservation and access.

In summary, full-reference metrics such as MSE, PSNR, VDP, VQM, and SSIM play an essential role in evaluating the quality of digitized images of cultural heritage. These metrics, together with ISO 19263-1 and ISO 19264-1 standards, ensure that digitization processes are rigorous and consistent, thus preserving the richness and history of works for future generations.

Summary of chapter 3.11 of the book «Digitization and Preservation of Cultural Heritage»