This project is solving the Leaf Me Alone challenge. Description
Our project aims to automatically detect leave injury from leaf pictures taken from a mobile camera. The images must have a uniform background (say a leaf agsint a sheet of paper) for easier segmentation. Segmentation is achieved by clustering the pixels is leaf/no-leaf clusters, using the color space with Expectation Maximization (EM) algorithm.
Once segmentation is done, we know the total area of the leaf pixel-wise. Then we apply super pixel algorithms to detect components within the previously segmented leaf area. These components will have average pixel colors, which we can then cluster again with EM to define 2 new clusters: Injury / No Injury. Then we take the components with colors close to green as good components, making the others as injury.
Once injury components have been detected, we calculate the percentage of injury based on the total area of the leaf, providing the user with the damage percentage. Additionally we detect the geo-location where the picture was taken, and compare it against an ozone database, to provide the user with additional information about the injury of the leaf.
The data recovered from the user is also saved in a database for future reference, particularly for scientist who may find a crowsourcing approach viable for future ozone/injury research.
License: Academic Free License 3.0 (AFL-3.0)
Source Code/Project URL: https://github.com/maeotaku/O3_BackEnd.git
Front End - https://github.com/tzamora/LeafMeAlone