An image processor written in C#. Applies math operators on a given image set. Results in a set of data parameters to use for neural network training.

I developed this at university to transform greyscale image datasets of 19x19 pixels to a 7x7 image using a set of mathematical operators (Sobel, Kirsch, Prewitt, Scharr and Isotropic).

The image is downscaled by moving a 3x3 operator along the given bitmap, multiplies against it, sums up the 9 values and returns that as a single paramter for a NN to train with. It then continues to move along the image until all pixels have been transformed.

Each transformed face is represented as a single 50 column row in the output CSV file. The 50th column specifies the class (0 non-face, 1 face).

• gui interface
• live preview window (cool effect)
• multiple types of math operators
• multi-threaded
• displays image processing rate
• shows a realtime progress bar
• shows the time elapsed and total processing duration

For all files in this repository that don't contain explicit licensing, the MIT license then applies. See the accompanying LICENSE for more details.