Computer Vision:

Computer vision (also known as machine vision) is the construction of explicit meaningful descriptions of physical objects or other observable phenomena from images. Computer vision, the focus of the VIA group's research, has a very wide range of applications from medical diagnosis to seeing robots including autonomous vehicles, from particle physics to geological surveying, and from missile defense to quality control and basic scientific research. Wherever images play an important role in understanding a problem, there is a potential application for computer vision. A subset of topics in the field of computer vision include image segmentation and formation, edge detection, region growing and shape description.

Recently there has been intense interest in the application for deep learning methods to the computer vision task, especially in the context of search and social media businesses such as Google and Facebook. These methods are designed to automatically extract useful information from images (photographs) and videos that have been uploaded to web systems by users.

In general, in the deep learning method, a “neuron” learning network is “trained” on a very large number of image examples with known desired outcome responses. Using machine learning optimization algorithms, it associates the desired outcomes with the images, by passing through the training data set many times. Thus, it is the network that “learns” how to solve the task. This is in contrast to more traditional approach in which the system is designed to solve a task by analysis of relevant image features; that is, the solution to the problem is determined by engineering design. In general, far fewer training samples are needed to develop the traditional approach compared to the deep learning approach. There have been a number of recent computer vision applications in which the deep learning approach has been shown to outperform the traditional approach; especially in the domain of internet photographs. Consequently, there is currently a great deal of research activity to (a) better understand how to design the deep learning networks and (b) to apply the deep learning method to different computer vision application areas.

Relationships to Other Fields:

Computer vision is closely related to a number of different fields, yet there are clear distinctions between them. The goal of computer vision is to extract information from an image. That information may be a geometric description of an object, the distance between the object and the source, whether an airplane is an enemy or ally, or whether a patient has cancer.

The related field of image processing transforms an image into another image. The output image may be compressed, enhanced, or have corrections for blurring from motion. Often these algorithms are useful in the early stages of computer vision it is helpful to preprocess images to get them ready for computer analysis.

Computer graphics generates images from a mathematical description. In a sense, this is the opposite of computer vision, which returns descriptions of a scene based on an image. Computer graphics is image synthesis, while computer vision is image analysis. Other fields which are tightly knit with computer vision are pattern recognition, statistics, and psychophysics.

Medical Image Analysis and the VIA lab:

The current primary research goal of the VIA group is the development of a medical computer assisted diagnosis (CAD) system. The focus of this research, in collaboration with the Mount Sinai Medical School, is to evaluate patient health and diagnose disease (especially cancer) through analysis of three-dimensional images from CT scans. Current programs are directed towards detecting pulmonary nodules in whole lung scans and the diagnosis of nodules from high resolution CT scans. Research projects address the following image analysis tasks: Advanced 3D image segmentation, 3D visualization and animation, medical database developed, computer medical diagnosis methods including neural networks.

Click here to see a list of Current Research Projects.