The aim of our research is to assess image quality of prints without the involvement of human observers. The printing industry is continuously moving forward as new products and technologies are introduced to the market. The need to assess the quality is increased with this growth, for example, to verify that these technology advancements lead to higher quality prints.
In this thesis we describe the work carried out to use image quality metrics for the evaluation of printed images. The intended behavior of such metrics is to measure or predict image quality as human observers would perceive it.
Following an introduction and background on quality assessment, we introduce the concept of image quality metrics. Existing image quality metrics are classified and a survey of them is given, to show how they are constructed and their differences. Following the survey, a new image quality metric, the Spatial Hue Angle MEtric (SHAME) is proposed, which accounts for two key aspects of image quality, namely region of interest and the human visual system. The evaluation of image quality metrics against the percept is a key aspect for ensuring that the metrics can substitute or assist human observers in the assessment of quality. Therefore, existing evaluation methods are presented and analyzed, revealing the need for a new method to assess the overall performance of image quality metrics. For that reason, a new method to evaluate overall performance is proposed, based on the rank order method. Using existing methods and the new evaluation method, a set of commonly used metrics are evaluated with a set of public databases. These databases contain digital images, with a range of different distortions and quality issues, and have quality ratings from human observers.
The knowledge gathered from the evaluation of image quality metrics on the existing databases was then applied to the assessment of printed images. Since the metrics require digital images as input, a framework to digitize printed images is proposed. Using this framework a set of metrics is evaluated against human observers, which shows that none of the existing metrics predict overall image quality adequately. These findings lead us to break overall image quality into parts, more precisely quality attributes. Based on existing quality attributes a manageable set of six color printing quality attributes is proposed. The final set included: sharpness, color, lightness, contrast, artifacts, and physical. Through two experimental validation procedures these attributes are found to be a good foundation for the evaluation of color prints. The image quality metrics are then used to describe each of the proposed quality attributes separately to find the most appropriate metrics to measure the quality of each attribute. Two experiments with human observers were carried out, which acted as the basis for the evaluation and selection of metrics. The results show that for some attributes, such as sharpness, suitable metrics can be found, but additional work is needed to find metrics that correlate well with the percept for all of the attributes.
An area that may be improved with the use of image quality metrics is the reduction of quality values to a more manageable number (pooling), usually a single quality value. We have investigated the impact of pooling strategies on the performance of image quality metrics. This investigation shows that the performance is linked to the metric, and that the parameters for the pooling strategies are very important. Even with the effort spent on pooling strategies none of the evaluated metrics performed well for the color quality attribute. This lead to a proposal for a new image quality metric designed for the color attribute, Total Variation of Difference (TVD) metric, which applies a spatial filtering to simulate the human visual system before quality is calculated. A comparison against the state of the art metrics shows an increased performance for the new metric.
Lastly, we gather the work carried out in this thesis to develop a practical tool for the printing industry to assess the quality of prints using image quality metrics, named the Quality Assistant. The Quality Assistant consists of all functions needed to evaluate quality, including: a test image suite, the framework for digitizing the prints, a set of image quality metrics, and visualization tools. Through the work carried out in this thesis we have shown the applicability of image quality metrics for the evaluation of printing workflows.