The name of our web app, Braeburn, is named after a lightly colored red apple that was once used with green Granny Smith apples to test for colorblindness. We were inspired to create a tool to benefit individuals who are colorblind by helping to make public images more accessible. We realized that things such as informational posters or advertisements may not be as effective to those who are colorblind due to inaccessible color combinations being used. Therefore, we sought to tackle this problem with this project.
What it does
Our web app analyzes images uploaded by users and determines whether or not the image is accessible to people who are colorblind. It identifies color combinations that are hard to distinguish for colorblind people and offers suggestions to replace them.
How we built it
Challenges we ran into
One challenge we ran into is handling the complexity of the different color regions within an image, which is a prevailing problem in the field of computer vision. Our current algorithm uses an api to perform image segmentation that clusters areas of similar color together. This allowed us to more easily create a graph of nodes over the image, where each node is a unique color, and each node's neighbors are different color regions on the image that are nearby. We then traverse this graph and test each pair of neighboring color regions to check for inaccessible color combinations.
We also struggled to find ways to simulate colorblindness accurately as RGB values do not map easily to the cones that allow us to see color in our eyes. After some research, we converted RGB values to a different value called LMS, which is a more accurate representation of how we view color. Thus, for an RGB, the LMS value may be different for normal and colorblind vision. To determine if a color combination is inaccessible, we compare these LMS values.
To provide our color suggestions, we researched a lot to figure out how to best approximate our suggestions. It ultimately led us to learn about daltonizers, which can color correct or simulate colorblind vision, and we utilize one to suggest more accessible colors.
Finally, we ran into many issues integrating different parts of the frontend, which ended up being a huge time sink.
Overall, this project was a good challenge for all of us, given we had no previous exposure to computer vision topics.
Accomplishments that we're proud of
We're proud of completing a working product within the time limits of this hackathon and are proud of how our web app looks!
We are proud of the knowledge we learned, and the potential of our idea for the project. While many colorblindness simulators exist, ours is interesting for a few reasons . Firstly, we wanted to automate the process of making graphics and other visual materials accessible to those with colorblindness. We focused not only on the frequency of colors that appeared in the image; we created an algorithm that traverses the image and finds problematic pairs of colors that touch each other. We perform this task by finding all touching pairs of color areas (which is no easy task) and then comparing the distance of the pair with typical color vision and a transformed version of the pair with colorblind vision. This proved to be quite challenging, and we created a primitive algorithm that performs this task. The reach goal of this project would be to create an algorithm sophisticated enough to completely automate the task and return the image with color correction.
What we learned
We learned a lot about complex topics such as how to best divide a graph based on color and how to manipulate color pixels to reflect how colorblind people perceive color. Another thing we learned is that t's difficult to anticipate challenges and manage time. We also realize we were a bit ambitious and overlooked the complexity of computer vision topics.
What's next for Braeburn
We want to refine our color suggestion algorithm, extend the application to videos, and provide support for more types of colorblindness.