Parenting Handbook of Image recognition

Recently came a lot of younger brother are immersed in the image, but are not clear the clue, under the guidance of the teacher confessed to ask for the task. Here I've summed up some experience, or parenting manuals :

A image recognition is like teaching a child's argument, explaining to him the main characteristics of the goal; (try to use the characteristics he knows, or else you need to explain the definition of each feature)

For example, to identify a rectangle, you can tell him that the rectangle is made up of 4 straight segments (because he learned the Hough transform), the intersecting segments are perpendicular to each other (the slope is negatively reciprocal) ...

B, sometimes feel too many characteristics of the retreat and second, the interpretation of the target and the visual field of the other objects of the difference is good;

For example to identify a vehicle, you need to teach him what a car is.

If only the vehicle in the scene is moving, then just tell him "moving is the vehicle", as long as it explains what "moving" is, or what is "motionless" or "stationary" or "no change" (consider giving him the "frame difference" textbook, or the Advanced "optical Flow Method") He will understand, At least right in the current scenario ...

C, the other child at present mathematics (objective) study good, the language (subjective) particularly poor, in order to let him learn more efficient, the best use of mathematics to describe, lest also give him remedial language first;

Or vehicle identification, if the scene is very monotonous, only the ground with the car, then you will say "that is the car."

The word "that" is very cumbersome to explain, and it needs to be matched to the specified direction. You can say, "He sees the object is the car", so you just explain what is "object", such as the object has "shape" is also a "contour" (edge detection, he will be many methods Sobel, canny not too basic).

D, he will be very obedient, but also very serious, or particularly sensitive;

For example, you think that there is only ground and car, but he can find a lot of small "objects", such as the small stone on the ground, the shadow of the car. You can make him insensitive (vague or filtered, eg. Mean, median, Gaussian, and set the threshold for brightness); reinsurance Some, you may also tell it that only the "big" object is the car. The word "big" is rather vague, fortunately tuition is not too difficult-because "big" is relative, if you do not want to explain what is "small stone" words, you'd better give him a known reference, such as his "field of view", and finally you tell him that "the object that occupies your vision 1/3 is the car" ...

E, he can have some foundation, but most of the situation is in a field \ Scene we must start from zero;

Therefore we need patience and trust, because he is a genius, and because he is a genius, if he did wrong, the only possibility is that we teach the wrong, fortunately we can again and again.

F, in fact, not only image recognition, machine learning and even artificial intelligence, we can treat the computer as a good child ...

You can teach him by some description/feature, or you can use training to get him to explore--the more he knows, the closer he realizes the truth (but it's often hard to know whether his perception of the truth is exactly the same as ours, or even that he might give a current enough self-consistent truth from another perspective).

Parenting Handbook of Image recognition