Local Image Masking Database
The CSIQ local image masking database is a large database of visual detection thresholds for simulated localized distortion of natural images. Via a spatial three-alternative forced-choice procedure, we measured contrast thresholds for detecting a log-Gabor noise target placed within each local region of 30 natural-image masks. This database is ideal for testing models of visual masking.
If you use this database, please cite the following paper: M. M. Alam, K. P. Vilankar, D. J. Field, and D. M. Chandler, "Local Masking in Natural Images: A Database and Analysis, Journal of Vision, Vol. 14, 22, 2014. doi:10.1167/14.8.22.
- Target: The target was a 510x510 (11.7 degrees) log-Gabor noise pattern. It is zero-mean and ranges from -1 to +1. Download the target in .mat format here: err_img_sc4_or1_sine_phase.mat (2 MB)
- Mask: The mask images were 510x510 cropped versions of the images from the CSIQ image quality database. Download the cropped versions here: gray_masks.zip (4.5 MB)
- Thresholds: The local detection thresholds (in dB) can be downloaded here: Online_db4.30.2014.xlsx (128 KB)
- Info: The subject and image info are here: subject and image name info.txt (1 KB)
The conversion from pixel values to luminance for the display used in the experiment (Dell Trinitron P1130 Monitor driven by Bits++) was modeled as:
L = (b + k×P)g
where, b = 0.0794, k = 0.028, g = 2.358, P = pixel intensity (0 to 255), and L is luminance in cd/m2. We measured the luminance of our display for varying pixel intensities by using a Konica Minolta Chroma Meter (CS-100A). The luminance conversion parameters (b, k, g) were derived by fitting the luminance versus pixel-value curve by using the Matlab curve fitting toolbox.
Blocks and thresholds
Each of the mask images were divided into 36 patches (each of 85x85-pixels or 1.9 degrees). The target image was also divided into 36 patches. The contrast detection threshold for each spatial location was measured. For each of the 36 spatial locations for each image, we had a total of six threshold estimates (three subject per image per patch, and two runs per image per patch). Our contrast metric was RMS contrast [Moulden, B., Kingdom, F. A. A., & 943 Gatley, L. F. (1990). The standard deviation of luminance as a metric for contrast in random-dot images. Perception, 19, 79–101.]