After two years of work, the new camera appears with three major changes:
Detailed drawing
In addition to the dead-leaves field with high contrast is now also a DL field with low contrast (measurement of the fine drawing) on the test panel. As before, we determine the resolution of the colored structures for a border contrast of 50%. The limit of the resolution is reached when the resolution curve (image contrast) drops to 50% of the input contrast (MTF50). The low-contrast structures are now a direct measure for the preservation of fine structures without maximum contrast in the image. Unlike the old DL-direct method, we can now calculate not only the noise, but all artefacts from the frequency curve - DL cross.
Artefacts
In the course of image optimization, the camera calculates artefacts as additional and thus incorrect structures. We compare the camera image with the DL image on the test panel. Thus, the software recognizes which structures are maintained with which contrast and which have been added as artefacts afresh. This results in a clean DL value and provides a measured value for artefacts.
Edge lift
All cameras optimize the edge image so that the image appears sharper and more contrast-rich. The edge enhancement improves the resolution and DL values, which is also very useful when used moderately, but the picture is damaged.
Edge height
When a camera reinforces an edge, the recorded flattened rectangle curve is not the ideal rectangle curve of the original, but an intensified one with slight over- and under-oscillations, which appear as fine (ugly) parallel lines at edges in exaggerated use in the image. The calculation is based on the area under the over- and under-oscillators. We use the thus calculated edge values for high and low contrast to evaluate the corresponding DL and resolution results - this was done in the past by means of the curves and images Br>
Some cameras also increase the overall contrast of an edge. We record this as a percentage.
The following measurements remain unchanged
Here we go to the innovations in the test procedure in detail
The "Dead Leaves" test field consists of a random array of circles, which in turn have a random radius and a random color. The resulting pattern resembles the distribution of the spatial frequencies of a natural scene. If the distribution function of the position, radius and color of the circles is known, the power spectrum can be predicted. In the previous method, the performance spectrum of the template (known) was compared with the performance spectrum in the image (measured). Thus, for each spatial frequency, it can be determined whether and how well this frequency has been transmitted (DL direct).
The problem: Cameras do not only remove or reduce details in the image, but also add details through noise and artefacts. Only the noise we could consider over a gray reference surface, however, the noise reduction of the cameras on gray surfaces works differently than on colored structures. For other errors, however, we could not automatically recognize that the "information" was not from the original, but errors were the camera. In the first step, a camera with many artefacts got a good result for the transmission of the dead-leaf pattern. Therefore, in a second step, the curve had to be evaluated for the evaluation.
This problem is now solved by not only calculating the theoretical frequency content (power density spectrum) into the calculation, but also a theoretically ideal image of the original and the camera image (cross-domain density spectrum). Only the actual transmitted frequencies are now taken into account, which ignores artifacts like the noise (DL cross). This new process was developed by Image Engineering and presented at the prestigious "Electronic Imaging Conference" in San Francisco 2017.
Up to issue 11, a high-contrast dead-field was corrected by the noise components and the result was then evaluated manually. We now evaluate two dead-leaved fields (High Contrast and Low Contrast): A in the direct process, which results in artefacts as image details, as well as B in the cross-process, which leaves artifacts outside.
Artifacts added by the camera during signal processing result in a significant increase in the determined transfer function (green and purple curves on the right) in the old method (DL direct). The comparison with the new method (DL cross), which is not influenced by the artefacts, results in a measurement of the artefacts.
Optimized dead-leaf pattern
In this example, the Panasonic DMC-GH4 cuts significantly better in the old process than in the new one. A glance at the pictures explains this: At high ISO sensitivity, clearly visible artefacts are added to edges and fine structures.
Image Engineering's approach to the determination of the tripping delay has now also been written down in ISO standard 15781 and is therefore a worldwide standard. The camera to be tested is triggered by a mechanical "finger" and, among other things, an LED-based clock is used. Thus, the difference between "pressure on the resolver" and "recordings" can be determined very precisely.
The image is retrospectively sharpened by artificial over-elevation of edges in the signal processing. Here it is at the manufacturer a good measure to find. Without sharpening, a picture is often soft, too much sharpening leads to visible artefacts and an artificial imagery.
Note: It is much easier to sharpen an image in the post-processing on the computer than to correct an already sharpened image. The sharpening is indicated by a light or dark hem along an edge. The example pictures show a Nikon D610 against a Canon 6D. The latter is sharpened considerably more. IE checks sharpening on two edges of different contrasts and determines the height and size of the added seams for vertical and horizontal structures (Edge high / low)
Two cameras with different signal processing: Left the Leica T, a camera that is very conservative with sharpening and noise reduction. On the right the Samsung NX3000: Here is a much stronger signal processing. According to the old measurement method, the Samsung got the better measured values (1015 LP / PH vs. 1396 LP / PH), although this is not obvious in the picture.
Artifact measurement: Systematic capture of image defects
Time measurement: time under control
Determine post-sharpening: phantom lines
Measuring Fine Drawing
According to a new method of measurement, we are able to pinpoint exactly what this is: the artifact measured value is significantly increased, and the sharpening (measured value edge sharpening) is much more pronounced with the Samsung camera. As always, the Samsung achieves the better measured value according to its higher sensor resolution in the new cross method with high contrast (DL high contrast).
It has to leave the Leica at low contrasts (DL low contrast), since this receives more fine details. The measured values reflect the observed: the Leica shows more details, while the Samsung with sharpening and taking artifacts changes the image impression.
No comments:
Post a Comment