Introduction
In this article, some information is to be given about a data set to be used with the causal image models given in the articles [1], [2], [3] and [4]. The data set is created from the NIST special database 19. The information about and the download links for this database are given in [5].
The Creation of The Dataset
In the creation of the dataset, the image operations have been performed using the Python packages “opencv-python” and “scikit-image”. The information and tutorials for “opencv-python” and “scikit-image” are given in [6] and in [7] respectively. The version of “opencv-python” used is 4.7.0.68. The version of “scikit-image” used is 0.19.3.
A sample image is first converted to greyscale. A thresholding is applied to the resulting image. “opencv-python” package is used in these operations. The skeleton of the thresholded image is obtained using “scikit-image”. The skeletons with dimensions greater than $28 \times 28$ are resized to $28 \times 28$ dimensions. An adaptive thresholding is applied to the rescaled images to get the final skeletons.
The Source Code
The source code for the Image class is given at the link [8]. The code for creating the skeletons of the character “J” is at [9]. The images of the character “J” from the NIST database are also given at the link [10]. The $28 \times 28$ skeletons of most of these images are at [11].
Conclusion
Some information has been given about a data set of image skeletons. The images are from the NIST special database 19 of handprinted forms and characters.
References
[5] https://www.nist.gov/srd/nist-special-database-19
[6] https://docs.opencv.org/4.x/d6/d00/tutorial_py_root.html
[8] https://github.com/SaffetGokcenSen/skeleton_NIST_special_database_19/blob/main/Image.py
[9] https://github.com/SaffetGokcenSen/skeleton_NIST_special_database_19/blob/main/4a_skeletons.py
[10] https://github.com/SaffetGokcenSen/skeleton_NIST_special_database_19/blob/main/train_4a.zip