Measuring and Managing Digital Image Sharpening

With the development of digital image collections, comes the opportunity to capture and manipulate captured object information not previously available with photographic methods. With this advantage, however, come a host of choices beyond selection of image capture hardware, and acquisition software. In establishing imaging practice for a particular institution or project, it is important to understand the influence of the various choices on imaging performance. While this need is well understood for color management, it is less often considered in the capture of image detail. Captured image detail, and the related visual impression of image sharpness are commonly manipulated during image capture by image processing aimed at ‘sharpening’ the digital images. We propose an approach for sharpening management within the framework of existing ISO standards by use of the spatial frequency response (SFR) as referred to a known or implied SFR aim. Introduction For image capture and storage systems, image quality requirements are often described in terms of the intended use of the digital image content. For the consumer the quality of an image will usually depend on the perceived degree of excellence of the viewed or printed scene, which is often compared to the memory of a particular time, place and event. There are two basic aspects of image quality; the subjective impression of a viewer, and the technical or design details of the product or service that are needed to satisfy the customer’s needs or desires. In this paper, we focus on the influence of imaging practice and digital image processing (software) on the capture of image detail. We start with the premise that the management of imaging performance requires a good understanding of what imaging characteristics are important, and a reliable way to make objective measurements of them. Our approach is similar to that used in the development of color management programs. If color accuracy is important, as it usually is for archives and museums, establishing a colorimetric objective (desired color encoding) is a first step. Successful quality assurance then requires consistent periodic measurement of performance against the acceptable color-tolerances. Without these two steps, the best color-profiling system may deliver variable or inaccurate imaging performance. As for color measurement, the development of procedures for managing the capture of spatial detail has benefited from the development of performance standards for other applications. For example, the understanding of the factors that influence captured image sharpness has been helped by shifting away from simple sampling rate in favor of current ISO standards for spatial frequency response (SFR). One operation that is a common part of the digital imaging path is digital image sharpening. Generally, any operation that is aimed at modifying the visual impression of image detail, or sharpness, can be call image sharpening. Perhaps the most common time to apply sharpening is during image editing. Many image acquisition software (driver) programs also apply similar spatial image processing operations. Sharpening selection options, however, are often ambiguously labeled, e.g., ‘soft-look’, ‘standard’. Adobe Photoshop software offers five different sharpening filter operations with a range of user interfaces. It is not surprising, then, that common selection, and evaluation practice for image sharpening is qualitative and subjective. This can lead to variability of performance and confusion when comparing different systems. We propose an approach for sharpening management by use of the spatial frequency response (SFR) as referred to a known or implied input SFR aim. This idea was previously addressed by MacDonald. A method for routine performance evaluation will be described, along with the required test targets and analysis. Interpretation of results from collection images, and measurements will be emphasized. What is the SFR? Slanted-edge analysis has been applied to the evaluation of digital camera resolution for several years. This method is based on the image (or system output) due to an input edge feature of high optical quality. Often the measured edge response can be taken as an estimate of the MTF of the system. In other cases, the output modulation is divided by the input edge modulation frequency-by-frequency to yield the measured system MTF. In this paper we will refer to a measured or idea edge-based MTF as an SFR. Spatial frequency response (SFR) is a curve that characterizes how an imaging system maintains the relative contrast of increasing spatial frequency detail. The input variable along the horizontal axis of the SFR curve is spatial frequency, increasing to the right. Higher spatial frequencies translate to more finely spaced details. The output response along the vertical axis is the relative fraction of transfer (preservation) of contrast from object to digital image by camera or scanner. Ideally, one would like to maintain sufficient contrast of low, moderate, and high spatial frequencies. This is reflected by the SFR curve remaining relatively high with increasing spatial frequency (i.e., along the horizontal-axis). A typical SFR plot demonstrating this behavior is shown in the highest SFR plot, corresponding to the rightmost image, of Fig. 1. Due to factors such as lens design, assembly and defocus, and camera motion, blurring of the image occurs. This progressively reduces the spatial frequency content in the digital image, and is seen as a reduction in contrast and merging the light spaces with dark lines. Spatial details ultimately become unresolvable because so little contrast exists between adjacent image areas. The spatial frequency at which fine detail is no longer detectable, either visually or by machine, is the limiting resolution. For many cases, this limiting resolution occurs at the spatial frequency corresponding to a 10% response level of the SFR. This value is consistent with historical treatments of resolving power and effective resolution over the past century. It is also correlates well Archiving 2008 Final Program and Proceedings 89 with proposed ISO software solutions for reporting summary measures of camera resolution under ISO 12233 edition 2. 4 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 Frequency (cyc/pixel) S FR right image middle image left image Increasing frequency G rups of lim ing relution S FR G rups of lim ing relution Figure 1: SFR plots (top) and associated images demonstrating image sharpness and limiting resolution Though limiting resolution is a reasonable summary metric for objectively reporting spatial resolution, it does have limitations, particularly in predicting image sharpness. For instance, using a 10% SFR criterion, all of the images in Fig. 1 have the same limiting visual resolution. This is indicated by the loss of text visibility at the fourth text grouping from the top in each case. Notice, however, the remarkable differences in image sharpness between the three images. The rightmost clearly has the best image quality. The higher SFR values at all of the spatial frequencies in the companion graph predict this. This is followed by the middle and left most images with decreasing image sharpness, but equivalent limiting resolution. Figure 1 illustrates the importance of focusing on the low to middle spatial frequency range for the measuring and predicting of perceived image sharpness and overall quality. This fact has not been lost on the image processing community and is the region where digital sharpening operations are generally beneficial, when applied in moderation. Sharpness vs. Sharpening It is often said of digital imaging that sampling is not resolution. 5 Image sampling indicates the interval between pixels on a particular plane in the scene (camera), or on the object (scanner). Limiting resolution refers to the ability of an imaging component or system to distinguish finely spaced details. Although image sampling (e.g. 300 ppi vs. 600 ppi scanning) can enable a level of detail in a digital image, it is not the same as, and does not guarantee, the capture of a particular level of limiting resolution. High image sampling is a necessary but insufficient condition for resolving detail. Likewise, high (perceived) limiting resolution does not guarantee an overall impression of high sharpness in a displayed image. This was shown recently in Fig. 3 of Ref. 5, part of which is reproduced in Fig. 2. This graph shows the measured spatial frequency response, of the two image capture paths from digital still cameras. The results are based on the standard analysis of an edge feature in a sample image from each camera. The differences in the solid and dashed black lines at high frequencies help explain the perception of limiting resolution for the two systems. The frequency at which the SFR falls to 10% is indicated as the measure of limiting resolution. The system responses in the lower frequency range, 0.1-0.2 cy/pixel correspond to the differing impression of image sharpness from the two systems. Figure 2: Measured spatial frequency responses for two digital camera paths. Unsharp capture followed by digital sharpening (solid line), and well-focused optical capture without sharpening (dashed). From Ref. [5]. As indicated by the caption for Fig. 2, the camera image corresponding to the solid line had been subjected to an image sharpening operation. Many digital cameras and scanners apply such image processing operations as a routine part of image capture. These operations can take many forms, but all aim to enhance certain important image content. Sharpening image processing operations operate on a digital image after capture, and so do not completely compensate for, e.g., poorly focused optics, but can be useful in improving the appearance of an image after capture. Sharpness is a visual attribute of a displayed image, and there are image quality models which attempt to predict the level of sharpness that a viewer would perceive. Understanding both image sharpening operations and sharpness models can be done using the spatial-frequency description provided by the system SFR. 0.5 1.0