aeropuerto / img2signal

Image analysis tools for research, runs in your browser

img2signal

The goal of this project is to create a library of tools for analysing images inside a browser using HTML5 (JS & Canvas). This project is mostly meant for academic purposes and to aid research in the field of image manipulation & convolutional neural networks.

Setup

Clone the repository, open index.html in your browser. It should be noted that only Firefox will work with the initial commit as Chrome will not allow to process imageData from a canvas with images loaded locally without overriding security settings (not recommended). An improvement is on the way that will allow more convenient running, development and playing around.

Usage

Some example images are provided inside ./img, which you can toggle on and off by commenting stuff in/out in index.html. The main -method loads an img -tag with id="source" and processes it according to the given setup in main () inside img2signal.js. Feel free to play around with the setup in main depending on your needs.

Usage examples

• Toggle between image types of a simple image like ./img/red.jpg and ./img/red.png, notice how their signals differ in regards to noise
• Play around with different impulse responses (1D and 2D) for convolution, craft your own filters and see how each impulse response manipulates the signal and the resulting re-composed image.

Version history

0.2 Convolution

• Add support for 1D and 2D convolutions
• Add some example filters & impulse responses / kernels (e.g. gaussian blur, high pass)

0.1 Initial commit

• Support for plotting RGB channels of a given image

License

MIT

img2signal's Issues

There's no build pipeline

Add webpack.

Add webpack