• 

  The opening screen.

• 

  You point something cool, or make a collage of images, we find you some cool album artwork, and tunes to go with it.

• 

  (This is actually the image that was paired with the above!) A cool album cover and song that match the tone of the above image.

Drawdio

Upload or draw your image, and find albums with similar artwork. Find your mood music visually.

Our Mission

Album artwork is often unappreciated, but it can have a really strong connection to it's corresponding music. By focusing music discovery on the artwork, we've created a whole new way to explore, and connect with music. We hope to share that love of discovery and music with others!

The Algorithm

Finding images that are similar in feel is a difficult task, especially for a project with the explicit goal of using no neural networks. We wanted to have control of our algorithms ourselves, and that meant some clever tricks.

We used a mix of traditional and brand new methods of computer vision analysis to create an algorithm which aims to match images by their visual composition and feeling. Our goal is to make each image you give to us feel like it could sit next to the album artwork we pair it with in a gallery. To do this, we calculate a number of extremely complex metrics, and then ultimately compress them into a single n dimensional corpus vector which can then be queried spatially in a database to allow for incredible speed in a matching process that requires immense depth and thought.

Point Cloud Clustering Detection

For one of the metrics our vector corpus tracks, we decided that traditional rough edge detection and corner detection weren't enough to capture the information we wanted. In particular, we wanted to be able to determine the compositional layout of an image. In order to do that, we created a new algorithm for point cluster detection which helps us determine how an image is laid out really quickly!

Built With

• css
• html
• javascript