Resolution
In the SLR test, the laboratory measures the resolution, ie the smallest structure that can be displayed for the camera, for standard focal lengths and four ISO values from ISO 100 to ISO 1600. For compact cameras, we test the wide angle resolution for ISO 100, ISO 400 and the highest possible sensitivity as well as the Telephoto solution for ISO 100. In this case, we specify separate values for the center of the image and the corners to check the power fluctuations of the integrated lens. The test procedure was developed together with the chair for physical optics at the Cologne University of Applied Sciences. It uses a testchart with nine (SLR) resp. Five (compact camera) distributed over the image field siemensternnen. They are not ordinary ordinary stars, but those with sinusoidal modulation of the flanks, in which the black of the rays does not abruptly change into the white of the background, but decreases or increases according to the sinusoidal function. The testers photographed this chart with the respective camera under uniform laboratory conditions and evaluate the photos exactly using special analysis software. The decisive question is: how high is the contrast of the photographed structure, that is, the recorded Siemen stars, compared to the original? As a rule, the image shows a lesser contrast, which also decreases continuously with increasing fineness of the structures (higher frequencies) and thus toward the center of the star. In the Siemensstern, each radius corresponds to a certain frequency. When evaluating the test images, we check the radius at which the contrast of the image is only 10% of the original contrast, and the frequency associated with this radius is the finest structure still resolvable. The laboratory detects this so-called limit frequency for eight directions of each star. A score for the resolution results from these measured values. In the compact video camera, the video screens and the image center are used in equal parts. In the case of SLRs, on the other hand, we consider the performance in the center of the image to be 100%, since the edge drop is mainly due to the quality of the lens. In order to keep its impact on the results as low as possible, we select a particularly good macro as a test objective. This is how the ColorFoto has been tested since 2004 according to the revised ISO standard 12233. The measured values also show the contrast curve, which we print to each tested compact camera and lens.
Contrast curve of a compact camera
The solid curves represent the performance in the center of the image, the dashed lines for the image on the slabs. Green indicates the measurement results for wide-angle focal length and ISO 100, red the values for wide angle and ISO 400, blue the contrast for telephoto and ISO 100. The diagram shows: Low image resolution hardly results in a contrast. The finer the texture of the image (to the right), the weaker the contrast of the image until it finally falls below 10% and thus below the greased horizontal line - the interface indicates the cut-off frequency. This is below the theoretical maximum of the sensor, the Nyquist frequency (purple-colored line), which is calculated from the number of pixels in relation to the image height. The mountain of the contrast curves touches the horizontal 1.0 line in this camera: the contrast in the image remains unrestricted. The ideal case, the real cameras can not hold over the entire frequency range. The wider the "curve belly", the better. From time to time, we also observe values of more than 1.0 on rough structures. Digital cameras usually raise the contrast via internal post-processing to optimize the resolution. As a result, some models can even show a higher contrast in coarse structures than in the original (value greater than 1.0). However, such interventions are often associated with side effects, for example with unnaturally hard contours. The theoretical maximum resolution (Nyquist limit) can also be exceeded with a special pixel arrangement on the sensor. Reason: The Nyquist limit only refers to the vertical height. We also determine the resolution for four diagonal directions, where the resolution can be 40% above Nyquist. In this case, a real theoretical maximum, which is 20% higher than Nyquist, would result from the four raised values and the four values corresponding to the Nyquist limit. However, this potential is de facto exhausted by almost no camera, as most CCDs have only one color (red, green or blue) for each pixel and the missing colors have to be included. In addition, there are camera-independent filters that suppress image errors, but usually press the resolution below the Nyquist frequency.
Noise
Each digital photo shows a certain noise, so artefacts, which are not in the subject, only in the recording. A number of different errors are involved, one of which is caused by the sensor ("fixed pattern noise" and "thermal noise"), the other by natural irregularities of the light (photon noise) or by conversion to RGB images (color noise) Code>
The "fixed pattern noise" arises from the slightly different sensitivity of the individual pixels and manifests itself in the image by a grainy structure. This is a camera-specific weakness which every manufacturer tries to correct internally by means of white calibration.
The thermal noise is due to charges which do not occur as desired by light incidence, but by chance by temperature influences. The higher the temperature, the stronger the thermal noise. In order to achieve comparable results, the test laboratory always keeps the temperature at 23 ° C + -2 ° C. Apart from the temperature, the strength of the image defects depends on the exposure time and the pixel size: the shorter the exposure time and the larger the pixels, The better the result. The individual pixels "collect" more disturbances during longer exposure times. In addition, smaller pixels necessarily have a smaller light-sensitive surface, which means they are less sensitive to light and therefore deliver less charge with the same amount of light. The signal must be increased "artificially", which also increases unwanted noise.
Particularly critical in this context is photon noise. It remains absolutely the same, regardless of how large the individual pixels are. However, since smaller pixels provide only a weaker signal, the signal-to-noise ratio is particularly unfavorable and the image errors are more clearly visible. The same applies to high sensitivities, in which the camera also detects a weaker signal and thus also has to raise the noise more automatically - a problem that is more serious than in the lights.
We describe all these image errors with a value: the Visual Noise (VN), which is determined by means of a chart with 20 differently bright gray pixels. For this purpose, we take the chart under laboratory conditions with ISO 100, ISO 400, ISO 800, ISO 1600 and, if necessary, with a maximum adjustable sensitivity, then separate the noise disturbances from the actual image content and evaluate them. In doing so, we take into account the perception of the human eye, which is less sensitive to very fine details than to coarser structures, which can be recognized by the human being at low contrast.
In order to capture this special color-dependent frequency dependency of the eye, we also transfer the RGB image of the camera to the human perception-based XYZ color space, and from there to the "oponent space", which in turn refers to an investigation The dyes and the processing of the color signals in the eye. Then we filter the result with the so-called CSF (constrast sensitivity function). It describes the contrast from which the eye can see the structures (in this case the noise artifacts) of a particular frequency. We start with a 40 x 60 cm print at 300 dpi. Then we convert the image over the XYZ into the Luv color space, since this is particularly suitable for the observation of small areas and thus the picture noise. In the next step, we determine the standard deviation for all fields and weight the results, with the brightness L entering more strongly than the two color axes u and v
1,0 L + 0,852 u + 0,323 v
In this way we get a Visual Noise value for all 20 Grayscales of the Testchart. However, only the 16 most important, particularly important results are included in the discussion. In addition, we weight VN values above 1.0 VN higher, and take them high 1.4. Thus, we take into account the fact that a visual noise of up to 1 VN is hardly noticeable, 1 to 2 VN as light, but visible noise, and over 3 VN a really disturbing noise impression arises.
Texture reproduction In addition to resolution and image noise, the detail reproduction, ie, the fine-grained drawing or texture, has now also developed into a very significant quantity for the image quality. As the manufacturers are placing more and more pixels on the small compact cameras, and the noise is becoming an increasingly problematic issue, it has been possible to help with camera noise filters. However, they eliminate not only the unwanted artifacts but, as a rule, also fine details that confuse them with image errors. In order to record and evaluate this behavior with a concrete measured value, we determine the brightness difference of all neighboring pixels in the test recordings of a sample template, enter them as a frequency value into a table and display them graphically. The result corresponds to a Gaussian curve. A noise filter, which also eliminates image details with the noise, generally does not interfere with the image: fine, low-contrast structures are deleted with the noise, and contrasts rich in resolution are retained. This nonlinear operation causes neighboring pixels with small differences in brightness to be set to the same digital value - the image loses in the drawing. If, however, more and more adjacent pixels have the same value, this leads to a more pointed curve in our evaluation, since there are more often "0" values. The kurtosis (bulge) now increases and serves as a measure for nonlinear, image detail-canceling interventions of the signal processing. Conversely, small kurtosis values indicate a moderate interference of the noise filter and many preserved image details (texture). In exceptional cases, compacts provide completely "muddy" images despite good kurtosis values. In this case, the noise filter has linearly engaged and erased with the noise rough as fine structures. At the same time, however, the resolution, which is retained in the non-linear operation, which retains contrast-rich structures, is not affected.
Object contrast The object contrast (dynamics capture) generally describes the maximum contrast, ie the difference between the brightest and the darkest area in the subject. If the object contrast of the subject exceeds the contrast that can be detected by the camera, image details are lost on their images: lights are emitted, shadows are running - a problem that can be solved v.a. Often occurs during sunshine. In the case of slide films, there are approximately 8 screens between the brightest and darkest display area, 12 screens for negative films, and top digital cameras provide 10 f-stops, which is sufficient in most situations - provided the exposure is correct. To measure the object contrast of a compact camera, we use a supervision chart with circularly arranged gray scale wedge, which corresponds to a contrast of 10 stops (ISO standard 14524). In the case of SLRs, a clear-cut chart is used, also with a circularly arranged gray scale wedge - by illuminating it from behind, this chart covers a contrast of 13 apertures. The corresponding test images provide information on how the camera can automatically convert different brightnesses of a subject into digital values. In the evaluation, we determine the OECF (optoelectronic conversion function) shown in figure 4 separately for all three color channels red, green and blue. Ideally, the three curves are identical. At the same time, the object contrast of the aspirant results from the OECF, as the difference between the exposure leading to saturation (pure white surface) and the exposure required to produce a dark gray field, where the image noise is three times less Is called the signal. The object contrast detected in this way is generally indicated in densities or in diaphragm stages, with an aperture stage corresponding to approximately 0.3 densities. An object contrast of 1000: 1 is equivalent with 3 densities or 10 diaphragms. As a rule, the object contrast decreases with higher sensitivity, because the more the noise in the shadows, the lower the usable dynamics. White balance The camera electronics must perform a white balance so that an image is neutral in color at different light sources. The quality of this adjustment can be determined by the distance of the RGB values of a gray area: If the red, green and blue values match, the white balance and gray appear gray. We check the white balance under daylight conditions and the gray wedge of the OECF chart. In this case, a gray of medium brightness appears gray or neutral, but at the same time a lighter or darker gray has a color tinge. Color reproduction In order to determine how the compact camera processes colors internally or how exactly the colors of the original motif are used, we use the ColorChecker SG from Gretag Macbeth as a testchart. It is modeled on the colors of a natural scene. The measured values are clearly differentiated from those of other test labs: we not only determine the DeltaE color space for each color field, but also the difference in color saturation, color tone and brightness. The printed value indicates the mean deviation. Chromatic aberration (compact cameras only) For compact cameras, we also measure the quality of the integrated lens, or its imaging errors, including the chromatic aberration, which is noticeable in the picture with colored seams, especially on the slides. It goes back to the special property of optical glass or built-in lenses to break each wavelength differently. As a result, the green, red and blue suits are not exactly covered. For example, when the blue image is larger than the red and green due to the particular refractive characteristics of the objective lenses, blue color fringes appear. We measure the offset of the crosses of the distortion chart in the three image planes (red, green, blue) subpixel-accurate and at more than two hundred points in the image. Then we calculate the mean value from the ten largest differences between the three color channels. Vignetting (only compact cameras) Lenses form the corners darker than the center of the image. In order to express this so-called edge shading (vignetting) exactly in numbers, we illuminate a milk glass very uniformly using an Ulbricht sphere, photograph a uniformly light-gray surface with distance setting "infinity", and evaluate the differences in brightness within the resulting test pattern at 1200 measuring fields out. This results on the one hand in a meaningful mean value as well as in a graph illustrating the edge drop. For cameras with a zoom lens, the vignetting for wide-angle and telephoto is determined. Distortion (only compact cameras) In addition to chromatic aberration and vignetting, the distortion also belongs to the typical and often disturbing imaging errors of lenses - to be seen on actually straight lines, which curl at the edge of the picture. The rectangle pattern of the testchart does not remain a rectangle, but gets a barrel or pillow shape. The lab measures the number of lines on the edge of the image in relation to the total image height, and displays the result as a percentage value. AF + Trip Delay The test laboratory measures the autofocus speed including trip delay to 1/100 sec and 30 lux (moonlight, SLR only) and 3000 lux (daylight, SLR only) or 300 lux (illuminated room, compact camera only). The time that elapses from the pressure on the trigger to the recording counts. First, the camera is focused to infinity, which is a good approximation of a distance of 1000 times the focal distance. The tester then photographed a 1.5 m LED array. The first LED flashes at the same time as pressing the shutter button. The second, third to the 100th LEDs light up at a distance of 1/100 s. Then the recording shows which LED the camera actually triggered. Image Timeout (SLR only) We determine the possible number of still images per second for a typical JPEG image with maximum resolution and lowest compression.
No comments:
Post a Comment