Recognising objects is one of the things we do best, and it comes natural to us. On the contrary object recognition, by looking at their shape, has always been a challenge for computers (and for the software managing the computer vision...).
To tell the truth, we don't really know "why". Researchers have tried different approaches to recognise objects via software and although progresses have been made our eyes (and brain) is way faster and better in recognising objects in general situations.
Nevertheless works continues and new approaches (also leveraging on the enormous computation capability of todays microprocessors) are making computer vision and object recognition much better closing the gap between us and them.
Take a look at the graphics. There you'll see a comparison of three software programs accuracy (based on deep neural networks) in detecting an object side by side, and a reference to our human ability to detect an object in a similar situation. We -still- win.
Interestingly, as the light condition gets worse, no longer perception of colours, perception of the silhouette only, we are way better than those programs. This just means that our brain is able to identify an object using less clues than a computer.
This comparison was made by three researchers from the University of Leuven and are published on PLOS, Computational Biology in an open paper that makes for an interesting reading.
Notice that, as it happens for humans, the programs are not "programmed" to recognise shapes and objects, they learn to recognise them as they get more and more experienced!
I wonder for how long time we, humans, will outperform computers in this area, one of the few where we still have the upper hand....