Computer Vision News - May 2016

Project Management Tip Our CEO Ron Soferman has launched a series of lectures to provide a robust and still simple overview of how to ensure that computer vision projects respect goals, budget and deadlines. Mr. Soferman will share one of his project management tips in each of the upcoming Computer Vision News issues. This month we are going to learn about: Variety of data COMPUTER VISION NEWS 20 two aspects of this issue deserve a closer look: on one side, variety of imaging conditions ; on the other side, variety of objects and scenes . Variety of imaging conditions covers matters like illumination, colors, texture, the different angles of the image and the devices used in order to grab it. For instance, the very large number of different smartphones available entails a wide check of all those devices the use of which we plan to allow. The typical scenario starts with checking Android and iPhones in their most popular models, to verify that the software can work with images taken with each one of them. The resulting database with images from all the cameras will be used as input during the validation phase . The same is true in the medical area, with the exception that medical images are taken with more sophisticated devices like CT and MRI scanners , which are produced by major manufacturers like Siemens, GE, Philips, Hitachi and Toshiba: of course, the more popular a device is in a specific market, the most critical it will be to include its images in the study. The same is true for ultrasound and X-Ray scanners . Regarding variety of objects and scenes, the well known example is face detection software : you have to take into account all possible colors of faces, to make sure that all ethnic origins are correctly recognized by the software. Disregarding a specific color might prevent the system from recognizing the individual. Of course, when we talk about variety of objects in the medical area, we think at the different pathologies. It is very easy to develop a software which measures and detects features in a healthy person, but when you inspect pathologies the range of possible phenomena can be very large. Developer has to check very carefully what the assumptions are and when they are valid or not. Intentionally malicious activities must also be taken into account: a user might be trying to check the limits of the system or to overcome its expected functionalities. A typical fraud example might be that of somebody’s identity being stolen using an image copied from social media accounts. In conclusion: the test procedure has to include all the variety of data and to create a database as rich and diverse as possible, consulting field experts such as radiologists (for medical imaging), cytologists (for microscopy) and so on. “ the test procedure has to include all the variety of data and to create a database as rich and diverse as possible, consulting field experts ” Variety of data is one of the most common problems in computer vision projects. At least