Image Enhancement

Enhancing an image means modifying one or more of its visual features in order to increase the visibility of its content and enable its understanding. This is very important both in human as well as in machine vision. Algorithms removing or lowering noise, shadows, artifacts or color casts due to the illumination under which an image has been acquired are highly relevant to a plethora of applications, including for instance robust object/people detection and tracking, analysis of medical images, image rendering for entertainment and augmented reality, 3D reconstruction.

Image Enhancement Datasets

The page lists some datasets of real-world color images, acquired by different devices, in indoor and outdoor environments, under different light conditions, including low-light and back-light. Many images are characterized by the co-existence of bright and dark regions with different size and proportion. These features make these datasets particularly appropriate to test image enhancers based on spatial color analysis.


Datasets are stored as .tgz archives. You may use gunzip and tar to unzip them.


Datasets


  • SCA-30: this data-set includes 30 pictures in bmp format, named sca-30 (1).bmp, …, sca-30 (30).bmp. These images have low resolution: their size was set up on average to 44M in order to make the data-set adequate also for testing and comparing image enhancers having a very high computational complexity.

[Size of compressed archive: 2.4 M]


  • Pict20: this data-set contains 20 images in jpeg format, named 001.jpeg, …., 020.jpeg. Their size ranges from 123M to 335M pixels about. The dataset includes also some images from SCA-30 but captured with higher resolution.

[Size of compressed archive: 616 K]


  • Pgshutter: this data-set consists of 20 images, acquired by the FLIR camera with auto-exposure off. Most images, that have size 640 x 480, are very dark. This makes the data-set adequate to evaluate image enhancement under low light conditions.

[Size of compressed archive: 324 KB]


  • MEXICO (Multi-EXposure Image COllection) - this data-set contains 400 color pictures with size 640 x 480 depicting 40 scenes of real-world indoor or outdoor environments characterized by challenging issues for image enhancement, as e.g. the co-existence of dark and bright regions at different proportions, back-light, shadows, chromatic dominants of the illuminant, presence regions with different granularity, from uniform to highly textured. Each scene has been captured with 10 different exposure times (from 3 to 30 ms with step of 3ms) by the FLIR Camera. The images, which are saved in jpeg format, are organized in 40 folders: the folder X contains the 10 pictures of the scene X. The images of each folder are named Img_Shutter_Y.jpeg, where Y codifies the exposure time at which the image has been acquired (Y = 03, 06, …, 30 ms). The data-set MEXICO has been used in ref. [1].


  • MEXICO-gr20 - This data-set is a subset of the data-set MEXICO including 60 images. MEXICO-gr20 includes 20 scenes, each of which represents an indoor or outdoor environment acquired with the FireFly camera under the three exposure times 6 ms, 21 ms, 30 ms. These images, originally acquired as color pictures and with size 640 x 480, have been half-scaled and converted to gray-level. The suffix “gr20” in the data-set name is just an acronym reminding that the images are in gray-level (“g”), re-sized (“r”) and their number is 20. Each image is named ‘sceneX-Y.pgm” where X identifies the corresponding scene and Y codifies its exposure time (Y = 06, 21, 30).

The size of the data-set is 4,5 MB, the archive downloadable here requires 3,0 MB of space.

The purpose of the data-set is to allow the tests of algorithms working on a single image channel and characterized by high complexity and long execution time, such as spatial color algorithms or image descriptors processing the image pixel-by-pixel, The multi-exposure times (corresponding to under-, well-, over- exposed images) enables to investigate the robustness of image processing methods to light dimming (thus to shadows or changes of the intensity of the light of the acquired scene), and to test HDR methods as well.

images from the dataset

MEXICO example - shutter speed from 3 ms to 30 ms


Software


Along with the datasets, we provide also the following software, that is provided as a .tgz archive along with license and instructions:


  • GREAT-v2019: an implementation of the spatial color algorithm GREAT[1] [2] as a binary executable file running on Windows 10 x64.


References

  1. M. Lecca, A. Rizzi, R.P. Serapioni, GREAT: a gradient-based color-sampling scheme for Retinex, Journal of the Optical Society of America. A - Optics, Image Science, and Vision, vol. 34, n. 4, pp. 513-522, 2017

  2. M. Lecca, Michela, A. Torresani, F. Remondino, On Image Enhancement for Unsupervised Image Description and Matching, International Conference on Image Analysis and Processing - ICIAP, LNCS vol.11752, n. Part II, pp. 82-92, 2019