Superresolution testing by MTF measurement

Introduction

MTF50 (MTF stands for Modulation Transfer Function, also known as SFR - Spatial Frequency Response) is the common parameter used for measuring image sharpness - the most important image quality factor.
This article describes the MTF comparision of digital photographs, processed by PhotoAcute and the original photographs, taken with the same digital camera.
Tests show up to x2 MTF increase, which is equivalent to x2 resolution increase.

You can find more information about MTF on Imatest site.

To skip the description and proceed directly to the results, click here.

Download FREE trial of PhotoAcute Studio

The procedure

To measure MTF, one of the leading image analysing tools - Imatest was used.

The tests were performed with the photographs taken with Canon EOS 300D digital camera with EF-S 18-55mm f/3.5-5.6 USM lens (kit) and EF50mm f/1.8 II lens.

The sample target is the gray-white vertical edge, tilted 5 degrees. The photographs were taken in series of 8 frames in each and processed by two versions of PhotoAcute Studio: version 1.0, processing 8-bit JPEG images and the Pro version (which was in development at the time of creation of this article), processing 16-bit RAW images.


Sharpening

Since every digital camera sharpens images to some degree, the comparision of MTF between the sharpened image and the PhotoAcute procesing result (especially RAW processing result) may have little meaning. Strong sharpening may lead to high MTF50, but poor quality. Little sharpening (note that PhotoAcute does not apply standard sharpening, allowing user to perform it in specialized applications) will have an MTF50 that doesn't indicate real image quality.

To obtain a good measure of sharpness, the differences in sharpening must be removed from the analysis. Imatest tool has a special feature for removing the difference in sharpness - setting the sharpening of all images to a standard amount. This means sharpening undersharpened images and de-sharpening (blurring) oversharpened images.

In this test we used both ways - measuring MTF50 without standartized sharpening and with standartized sharpening. In the final results table, the results obtained with sharpness correction are indicated in parentheses.


Measurement: Canon EOS 300D, EF-S 18-55mm f/3.5-5.6 USM lens

Original photograph





Photographs, processed by PhotoAcute Studio (8-bit JPEG images)





Photographs, processed by PhotoAcute Studio (16-bit RAW images)






Measurement: Canon EOS 300D, EF50mm f/1.8 II lens

Original photograph





Photographs, processed by PhotoAcute Studio (8-bit JPEG images)





Photographs, processed by PhotoAcute Studio (16-bit RAW images)

Results

This table shows MTF50 measurement results for Canon 300D with EF-S 18-55mm f/3.5-5.6 USM (kit) lens.
In parenthesis: data for images corrected with standardized sharpening (see the detailed description above).

Image	MTF50, LW/PH	Effective resloution, MPix	Resolution increase, % of the original
Original image	1243 (1323)	2.32 (2.63) MPix
PhotoAcute Studio, 8-bit JPEG processing	1478 (1600)	3.28 (3.84) MPix	+41% (+46%)
PhotoAcute Studio, 16-bit RAW processing	1596 (1703)	3.83 (4.36) MPix	+65% (+66%)

This table shows MTF50 measurement results for Canon 300D with EF50mm f/1.8 II lens.
In parenthesis: data for images corrected with standardized sharpening (see the detailed description above).

Image	MTF50, Cycles/Pixel	Effective resloution, MPix	Resolution increase, % of the original
Original image	1459 (1450)	3.19 (3.15) MPix
PhotoAcute Studio, 8-bit JPEG processing	1678 (1824)	4.22 (4.99) MPix	+32% (+58%)
PhotoAcute Studio, 16-bit RAW processing	1993 (2433)	5.97 (8.9) MPix	+87% (+183%)

We thank Imatest for providing the evaluation version of their excellent performance testing software.

FAQ Correct me if I'm wrong, but an improvement should show in increased MTF50 and a general lift up of the curve rather then a falling one, shouldn't it? Not only close to MTF10 or the Nyquist-frequency but also in lower spatial frequencies. The discrepancy you've noticed is because of the PhotoAcute output being twice the size of input image. To get a more consistent number in Cy/Pxl value we could have interpolated the original image to the size of PhotoAcute output, but interpolation itself will contaminate the measurement. Trying to calculate the LW/PH I got confused. A standard formula for that (CY/PX x 2 x Number of vertical pixels (per PH) ) works well for the original test image with a PH = 2048. But when calculating the enhance images it seems either my equation is wrong or that there must be a different sized image with a higher PX/PH for those images. The reason is the same as for question above, the PH of output image after processing with PhotoAcute is 4096. The behaviour of the SFR in the lower spatial frequencies seems sometimes a bit odd. The overshoots over SFR = 1 showed in the charts labeled as enhanced by PhotoAcute reminds me of typical oversharpening from digital camera jpegs. The chart labeled as original looks more like a raw-converted image to me. Yes, the overshoot can be seen with red strike-through 'corrected' line. This line shows how the edge profile and frequency responce would look like if standardized sharpening were applied. This is the feature of Imatest, not PhotoAcute. The pure results obtained with Photoacute are depicted with black solid line and do actually display a decrease in overshoot that is seen in original image.

Developed by Almalence   ::   Design by A.Green