Painting and Computer Vision

Painting and Computer Vision

 

2011-6-2

 

 

Recently, I read a book that teaches people to paint and think like an artist. This book claims that it can give people who do not paint basic painting capabilities in a few days. In general, the author thinks that our brain has two ways to observe the world: l mode and r mode. r mode observes the details, the L model analyzes the observed details based on our life experience, so as to achieve the goal of understanding the external world.

 

The author believes that after long-term life training, the R model is under the L model, so that when drawing, our information is from the L model, that is, the processed information of the brain is obtained. In this case, our painting has an interesting phenomenon. For example, when you draw a face, the ratio of facial features is greater than the actual situation, and the eyes are above the head (the actual position should be in the middle ). Another example is that a variety of background images can effectively interfere with our judgment of the object's position and size. through some simple lines or symbols, we can "cheat" the brain. In l mode, we cannot get the correct object size and proportion, because the brain modifies the image for later work.

 

The author claims that to solve this problem, we need to discard the L model, let the brain work in the r mode, and then train the hand to faithfully draw the images in the r mode of the brain, the basic painting ability can be achieved.

 

An interesting method proposed by the author is to copy the painting, and the L model will be abandoned by the brain due to difficulties in identification, so as to enter the R model, then the training is done with hands at a very slow speed and drawn according to the image you see (note that it is not an object. Based on the author's experience, all students use this method to improve their painting level.

 

The idiot painters reported in discovery have proved this theory from another aspect. These idiot painters have lost the ability to distinguish and understand everyday objects for various reasons, including text, facial recognition capabilities, but also show amazing painting capabilities. Some people can draw the entire street view after visiting the city, without any details. Some people can record all the content after looking at the blackboard filled with text, just like a camera, but have no idea about the text on it.

 

Therefore, it can be seen that image rendering is actually a "bottom layer" capability, and most people gradually lose this capability because of the acquisition of the acquired recognition capability, you can only go through the "anti-recognition" Training Method again to enter the underlying mode. This process is like a reverse process of "Pattern Recognition" in computer vision. In computer systems, cameras can easily obtain all the details within the visual range, but it is very difficult to achieve human recognition.

 

We may be able to get some inspiration from this reverse process. Why are we unable to grasp the length, proportion, and other information accurately in the L mode, but we can easily identify words and objects, and distinguish between faces and even expressions. Whether we are a camera that keeps sampling during early childhood, and the brain saves countless samples for learning and classification, thus achieving the L-mode capability. How does the brain process images? Can the brain understand the learning process and results of the L model through these "reverse processes" and "spoofing processes" to help computers simulate this process, thus achieving or approaching human recognition capabilities.