|
Description
Information about this dataset
Data
Links to data
Publications
Publications prepared with this data
|
|
Funt et al. HDR Dataset
The following is a data set of images of 105
scenes
captured using a Nikon D700 digital still camera.
The camera's auto-bracketing was used to capture up to 9 images of
exposures with 1 EV (exposure value) difference between each in the sequence. The rate of capture was 5 frames per second. The exposure
range was set to ensure that in each set there would be at least one image with maximum digital count less than 10321. During
bracketing, the camera was set to allow it to adjust the shutter speed and/or the aperture setting automatically between frames in order
to change the exposure by 1EV. In other words, the f-stop setting was not fixed. All images were recorded in Nikon's NEF raw data format.
The raw images were then processed in two ways. The first was to create almost-raw 16-bit Portable Network Graphics (PNG) format
(lossless compression) images from the NEF data, one image per exposure value. We will refer to these 16-bit PNGs as the 'base images'.
The second was to create a set of HDR (high dynamic range) images from the base images.
Two sets of base images were captured for each scene. One set includes 4 Gretag Macbeth mini Colorcheckers positioned at different
angles with respect to one another. The second set contained images of the same scene, but without the Colorcheckers. Between taking the
two image sets, the camera was refocused and possibly moved slightly. For the first set, the focus was adjusted so the Colorcheckers
were in focus. For the second set, the focus was optimized for the scene overall. (In the cases when only the first set is available,
the Colorcheckers are cropped out to form the second set).
To create the base images, the raw NEF images were decoded using dcraw (more specifically
Windows executable dcrawMS.exe). To preserve the original digital counts for each of the RGB
channels, demosaicing was not enabled. The camera outputs 14-bit data per channel so the range of possible digital counts is 0 to
16383. The raw images contain 4284x2844 14-bit values in an RGGB pattern. To create a color image the two G values were averaged, but no
further demosaicing was done. This results in a 2142x1422 RGB image.
An HDR image was also constructed from each set of base images. The base images require alignment, which was done by the simple
Median
Threshold Bitmap approach (Ward 2003). After applying a 3x3 median filter to the base images, the Matlab function makehdr from the
Matlab Image
Processing Toolbox was used to combine them into one HDR image. To ensure the reliability of the pixel values, all base image
pixels having values greater than 13004 or less than 30 were excluded. The final HDR images may
vary slightly in size due to possible cropping at the boundaries of the images as they are aligned.
Measuring the Scene Illumination (Ground Truth)
The Colorcheckers were placed in the scene at a point where the illumination incident on them was expected to be representative of
the
color of the overall scene illumination. While all scenes contain some variation in the illumination color because of
inter-reflections, scenes that clearly have strong variations in illumination color were avoided. For example, a room with interior
tungsten lighting mixed with daylight entering through a window would be excluded.
The illumination chromaticity is determined by manually sampling the RGB digital counts from each of the 4 white patches from the
Colorcheckers of the base images. The brightest image from each set not containing any overexposed pixels (i.e., all RGB < 2^14-1)
anywhere within the Colorcheckers is used, and each measurement is the average RGB of the 3x3 neighborhood of a pixel near the center of
the white patch. Since the Colorcheckers differ in orientation, we obtain measurements of the scene illumination at 4 different angles
of incidence. Not surprisingly these measurements do not always agree. For the tests described below, the median of the illumination
chromaticities from the 4 Colorcheckers is used as the ground truth.
|
Index: (Attributs of the images in each set)
download
- File Name: set_name + "index.mat"
- Type: Matlab struct
- Attributes:
- + index: {set_name, cc, img, cc_align_info, img_align_info, bound_box, cc19_coord_xy}
- | - set_name: name of the set
- | + cc: list of Colorchecker images
- | | - name: image file name
- | | - shutter: shutter speed
- | | - time: time stamp
- | | - iso: iso value
- | | - aperture: aperture value
- | + img: list of images without Colorchecker
- | | - ... (same attributes as "cc")
- | + cc_align_info: aligment information of the Colorchecker images
- | | - shift: offsets
- | | - excluded_x: horizontal pixels to be excluded
- | | - excluded_y: vertical pixels to be excluded
- | + img_align_info: aligment information of the images without Colorcheckers
- | | - ... (same attributes as "cc_align_info")
- | - bound_box: coordiate of the bounding box of four Colorcheckers
- | - cc19_coord_xy: coordinates of the white patch of each Colorchecker
Base Images: (Divided into 15 parts, 1~2GB each)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
File Name: set_name + image_name + shutter_speed + aperture + iso + time
HDR Images: download
File Name: set_name
Measured Illumination: (White patch values of the 4 Colorcheckers of each set)
download
Image Filelist:
HDR image list: set names of 105 HDR images
download
Base image list 1: filenames of 105 base images >= 10321
download
Base image list 1: filenames of 105 base images >= 11000
download
Funt, B. and Shi, L.,
The Rehabilitation of MaxRGB,
Proc. IS&T Eighteenth Color Imaging Conference, San Antonio, Nov. 2010.
Funt, B. and Shi, L.,
The Effect of Exposure on MaxRGB Color Constancy,
Proc. SPIE Volume 7527 Human Vision and Electronic Imaging XV, San Jose, Jan. 2010.
|