Human Guided Machine Vision

A human-computer symbiosis can be better than either alone

Traditional image processing algorithms produce maps from satellite images rapidly, but with mistakes. These algorithms can confuse roads for rivers, buildings for natural formations, fit contours to shadows etc. The alternative is a tedious manual process of marking up geographical features — slow, effort-intensive and inaccurate.

But do we have to think of humans and machines as oppositional? Or could an appropriate symbiosis of the two be better than both — fast and precise like a machine, yet having the understanding, creativity, and context of a human being?

Our Human Guided Machine Vision (HGMV) system is just that — human-computer symbiosis! It's built with our software framework of fast image processing and machine vision algorithms. And its GUI employs novel user interface principles allowing fast interaction and live feedback.

For instance, if you were trying to mark a river in the map, you could just point to it with your mouse, while the algorithm concurrently detects and displays its predicted version of the river in real time. If say the algorithm misses a tributary or includes a road, you could guide it by pointing to the correct path. Once you are happy with the predicted result, a mouse click will create the final markings on the map instantly.


Our HGMV system can be used to achieve real-time, precise, contextual results in many application areas:

Optical Metrology

It is important in many industrial settings to measure dimensions of parts, and image or shadow based methods are frequently used. Our HGMV product for optical metrology can draw tangents, fit curves and geometric shapes, find boundaries, find intersections, and perform gemetric constructions, all in the HGMV paradigm. The user can then extract parameters such as curvatures, slopes, angles, radii and lengths.

Satellite Image Processing

Road networks detected from aerial or satellite images are important for GPS navigation throughout the world. Road features change continuously, and hence keeping maps up-to-date is not easy. Accuracy of road detection is also important — after all, you do not want driving directions to run you off the road! As described in the example above, HGMV can help fast, correct and accurate detection of roads, as well as help in other geographical surveying tasks such as detecting forest cover, water bodies, arable land, etc.

Medical Image Processing

Detection of tumors, cancers, lesions, fractures and scars on images and scans is an important aspect of medical diagnosis today. Even though automated classification of medical features has made good progress in the past two decades, human understanding cannot and should not be replaced. Instead of replacing radiologists, oncologists and other diagnosticians, HGMV helps by improving their accuracy, and reducing their work.


HGMV can help archeologists, geologists, environmental scientists, particle physicists and myriad other scientists who detect patterns from scans, photographs, and other visual data sources. Tasks such as pattern detection, classification, measurement, etc. can be semi-automated with intelligent humans in the loop, giving accurate results while reducing tedium.


Image processing is a very important aspect of the entertainment industry in general, and movies in particular. Image manipulation is used in studio photography, special effects, image and video restoration and many other tasks. Using HGMV, we can increase the speed and accuracy of digital effects.