rstreet85 / uncanny Goto Github PK

Java library to detect edges in images using the Canny algorithm with variable parameters.

Home Page: https://rstreet85.github.io/uncanny/

License: MIT License

Java 100.00%

canny-edge-detection canny edge-detection edge-detection-algorithm java maven travis-ci computer-vision image-processing digital-image-processing

uncanny's Introduction

Uncanny

A pure Java implementation of John Canny's 1986 edge detector, including a Gaussian filter. The algorithm accepts an image, converts it to grayscale, blurs it with a Gaussian filter, and then detects the edges within it. It does so by finding the 'edge' magnitude of each pixel with a bi-directional Sobel operator, then disregarding all 'weak' edge pixels unless they are directly adjacent to a 'strong' edge pixel (hysteresis). The Canny is elegant - despite it's age and simplicity, it is still the standard for edge detection, and readily lends itself to optimization. This program is intended to provide quick method of detecting edges and study the Canny method, as well as getting quick feedback while tweaking parameters for specific projects.

Tester Class Usage

Command-line arguments: -fileName -outputFileExtension

Code Usage

//Sample JCanny usage
try {
    BufferedImage input = ImageIO.read(new File(imgFileName));
    BufferedImage output = JCanny.CannyEdges(input, CANNY_STD_DEV, CANNY_THRESHOLD_RATIO);
    ImageIO.write(output, imgExt, new File(imgOutFile));
} catch (Exception ex) {
    System.out.println("ERROR ACCESING IMAGE FILE:\n" + ex.getMessage());
}

Example:

test/test1.png png

To-Do

Allow user to set the paramters for Gaussian filter & hysteresis for optimization purposes
Add Pratt figure of merit functionality, so that user can designate what edges they want, and determine how closely the output from parameters match.
Explore additional preprocessing methods, such as white balance.
Current edge image is smaller than original, add buffer to fix this

uncanny's People

Stargazers

Watchers

Forkers

suryasinghvijay breslov vinisauter sauravchatterjee

uncanny's Issues

README Helpfulness

Modify readme to provide the most useful information to users.

Clean Javadoc in Gaussian

The Javadoc for the class looks sloppy and should be cleaned.

Remove unused methods from ImageUtils

Many are completely unused

Test Image SIze

Christ on sale, why is the test image 54MB in size??!! New/smaller test image is required. The output of the operations is only ~300KB.

Uncanny/Jcanny Gaussian Blur Paramas

This method doesn't allow caller to pass parameters for the Gaussian blur, it needs to do so.

Rename old classes: Jcanny

Jcanny needs to be renamed uncanny

Tester: Println v log

Replace System.out.println on line 51 w/Logger.

Tester Javadoc

The Javadoc for Tester.java looks sloppy, clean up pls.

Unnecessary comments in Gaussian

Remove any non-essential comments from this class. If they're needed, make them very clear.

Clean ImageUtils Javadoc

The Javadoc for ImageUtils looks sloppy, make sure it is accurate and helpful.

Detected Edges: Offset from Image

Hello Robert!
Thanks for a nicely-structured implementation of the Canny algorithm.

I try to utilize your code to identify the borders of an A4 page pictured over a wooden table (picture is attached).

I use your hardcoded defaults for GAUSSIAN_RADIUS (7) and GAUSSIAN_INTENSITY (1.5).
I call the JCanny.CannyEdges method passing-in the parameters used in your test code (within Tester.java), i.e. numberDeviations=1 and fract=0.2.

The result image is attached. It includes the source-image, after being gray-scaled, and on top of it - the detected edge-points, in red.

You could see that (A4) page-borders are not appropriately marked.
The detected edges are shifted a bit in relation to page's actual location.

Could you please advise - how to fix this offset? It would be much appreciated.

Source:

Result:

Result - Zoom: