Almost everything, I think, is close enough to the answer! Though computer vision today has progressed much with algorithms for object detection, perception, face detection, etc we have much to go. Else we would have got autonomous driving vehicles common place, your supermarket shopping would be done by a robot, blind persons can walk freely along any place, and so on.

Perception & Intuition

Human vision can perceive better and faster in complex situations than computer vision

E.g., Semantic retrieval

Computer vision algorithms for semantic retrieval research is still evolving and when a complex query is searched, it completely fails where as human vision with intuition could interpret it better.

It is difficult for a computer to retrieve meaningful results for query of a photo or video clip of Gore driving his car

http://www.forbes.com/sites/xerox/2013/10/30/how-we-use-computer-vision-present-and-future/

Recognize objets when they have been partially occluded and the machine can only view a bit zone. The machine requiere very much knowledge, previously in that

case.

The occlusion can be due to shadows or overlapping with other objects.

Other problem can be the detection of motion when the objects in the scene have similar texture and homogeneus surface.

Any references?

Our vision is actually quite crude - we do not see that many pixels, but we have a fantastic processor, which, remembers the picture and updates it with the natural small movements of the eye via interpolation, to complete the image. Furthermore, our brain can instantly recall and associate familiar objects, in all the possible orientations, and instantly give us the answer. The message then passes through the hypothalamus on it's way and we get an image which is a mixture of the new information. history and emotion (Is this object good, bad or neutral ) One will have to talk to a specialist in the medical profession to get a better “picture” (pun intended) of how exactly this works.

Our dilemma with machine vision is, that if we want our machine to "see" better, we increase the camera resolution, increasing the data stream and load on the database / algorithm..

Here is an interesting paper on machine vision by: Thomas Dean Mark A. Ruzon Mark Segal, Jonathon Shlens, Sudheendra Vijayanarasimhan, Jay Yagnik, "Fast, Accurate Detection of 100,000 Object Classes on a Single Machine"

Thank you all for the help! This is the first step for me.

We can focus a small area and the rest does not become relevant information. So our eyes optimizes images.

Allot, but here's something cool that is the other way around...

http://people.csail.mit.edu/mrub/vidmag/

As an extension to Mr Johnson's answer , we have a fast processor with neurons working at a good speed and ability to pull up information that are stored in the memory that help in decision making. Else the human eye is diffraction limited (resolution, depth perception...). Machine vision can be improvised and perfected by the algorithms and associated optics/electronics/systems .

I'd say that the main difference relies on processing rather than on potential. We do have the brain to process whatever we see, plus a lifelong training to understand the scene. Human vision works according to the paradigm of ACTIVE vision, i.e. we do not process everything at once nor with the highest resolution level. Instead, the human vision process has a strong (intention-based) feedback. We know about the size and shapes of objects via learning, so we can actually complete most of what our eyes are not perceiving due to occlusions and/or deformations. That's also why we are sensitive to optical illusions.

I think Jonas Gozdecki might have been referring to foveal vision (i.e. multirresolution images), which are handful when you want the highest resolution possible but do not have enough computing power to allow processing of a wide field of view (e.g. think of an eagle homing in on a prey from above to dive and kill). This is a good example of active vision: detection is performed at low resolution and then the highest resolution area is placed on the regions of interest in the image for detail processing.

In brief, I might be wrong, but if we had a computer capable of learning everything (about "seeing") that a human can, computer vision would be equal (or better) than human vision.

BTW, the book On Intelligence has a nice chapter of how human vision works at the brain and establishes some parallels with computer vision too.

Human vision performs well for all human life applications. It interprets 3D world from a sequence of images for these applications. From my point of view this is possible because human brain applies constraints based on its previous knowledge. But human vision is not perfect and for example it doesn’t perform well for rabbit life applications.

Computer vision performs better than human vision in some applications such as products quality control, guiding machines, process monitoring, etc. However, computer vision is far to solve human life applications.

Nowadays a good research challenge is to discover how human brain manages its knowledge and translate this method into a computer.

Human vision is good at managing abstract information, and owns enough knowledge that help much in daily life. Thus it's important to discover how the brain manages the knowledge and how they work in perception. Maybe deep learning would be a good choice.

Isn't it because that we humans already know what the object looks like, then we could predict what the object is, but in computer vision it does not know what the object is unless it has a prior knowledge (like the template)