OpenVL: A Developer-Level Abstraction of Computer Vision

OpenVL provides an abstraction at the task-level over complex computer vision algorithms. The abstraction is based on segments, distinct regions of the image defined by the developer through properties such as colour or texture (a) which are apparent in the image. Operations may then be performed on segments, such as pose detection (b), point correspondence (c) or face detection (d). For correspondence, a description of how segments vary between images is given; for detection, a template is defined for which groups of segments are matched. Computer vision is a complex field which can be challenging for those outside its research community to apply in the real world. Existing vision frameworks and libraries present APIs to users as lists of specific computer vision techniques (such as detection us-ing Haar cascades, tracking with particle filters, etc.). Application of these methods under real-world conditions requires deep algo-rithmic knowledge in order to (1) select the appropriate algorithm for the current problem and (2) to provide appropriate parameters to this algorithm to achieve a reasonable result. Requiring this level of expertise is a significant barrier to the widespread adoption of com-puter vision. We propose a task-level abstraction called OpenV