Automatic Contrast Evaluation for Android Themes

Several smartphone vendors now offer theme stores where one can submit themes that change the appearance of the OS. Theme designer must follow policies and guidelines to have their themes accepted. One common issue is the low contrast of elements in system applications, that can cause rework and delay to publish submitted themes. To prevent such problems, these devices need to manually check hundred of different screens. In this paper, we describe an automatic tool that walks in several application screens of a device running Android OS, analyzes them and generates detailed reports about low contrast issues. We use a shallow neural network that was trained with more than 10000 elements extracted from screenshots of several different Android applications. We show that our approach present high recall and precision when analyzing Android screen elements, and higher recall and precision when compared to low contrast check in the Google Accessibility Framework.