If you know something about face recognition, you will find that almost all face recognition applications are color, because of this, few people pay attention to whether the face recognition image is color or black and white.
Why does face recognition prefer color images?
This is the influence of the principle of human existence.
Roughly speaking, the principle of human existence is that the universe we see is like this, at least partly because of our existence. Hawking mentioned the concept of “the principle of human existence” in his book, which means: the reason why we see the universe like this is that if it is not like this, we would not observe it here. If all activities of scientific work are separated from the existence of human beings, they will not have any significance. The principle of human existence tells us that our understanding and description of the universe have our unique cognitive ability. The reason we see the universe is like this. It’s because if it’s not like this, we won’t be here to observe it, and we won’t ask this question.
Face recognition is similar to the principle of human memory in color images. The reason why we only recognize color faces in face recognition today is at least partly because we love the color world.
It is precisely because of this preference of human beings that almost most of the face recognition grabs color images, and the database stores color images. The algorithms all use color faces for deep learning. Also based on this preference, in places with poor light or at night, the camera uses white light instead of infrared light to get color images. Therefore, there is no chance for the development of a face recognition system for black and white images.
Does it mean that face recognition can’t recognize black and white images?
Of course not. It’s just that the recognition rate is lower and the error recognition rate is higher. But I believe that with the development of technology and the enhancement of deep learning for black and white images, this problem should be solved slowly. It just takes a certain amount of time.
Therefore, although face recognition is inferior to black-and-white images, it is not due to the algorithm design of face recognition itself, but due to the preference of data training.
It should be said that the development direction of face recognition system and technology cannot be decided by the system.
Color space applied in face recognition
Most Most of the current face image acquisition methods are color images. If we give up the color information, the information contained in the image will be reduced a lot.
When the light is different, the collected face images may be very different. The bright or dark light may change the collected RGB information, resulting in inaccurate recognition. At the same time, because RGB is the device related color space, when there are multiple devices or devices changing, the recognition failure will also occur. Therefore, before using the color information of face image, we must find a more stable component in the information.
At present, the mainstream face recognition methods based on color space all have one goal, that is, skin color separation. In most cases, skin color is considered to be the first extraction quantity in face recognition, and it is also a relatively stable parameter in color information. After skin color separation, the extracted contour and other information can be compared with the face data in the database more efficiently.
Difficulties of applying color information technology in face recognition
1. Color shift caused by ambient light
Different ambient light irradiation on the face will make skin color change, and the spectrum of the light source is different, so the spectrum of the reflected light cannot exceed the spectral range of the light source. In this case, the introduction of color technology may cause a reaction to the recognition results. On the basis of the accuracy of contour recognition, if the difference between the lighting source and the actual light source is too large, the final accuracy is lower than that of contour recognition. When the illumination condition is more extreme, the skin color of the face exceeds the threshold set by the machine, and it will be completely unrecognized.
2. The light intensity exceeds the threshold of the sensor
In most face recognition, the first step in the application of color information technology is to remove the brightness information from the color information. When the light intensity is too high or too low, the remaining color information is not accurate enough after removing the brightness information. When the brightness is too high or too low, the color information in the picture will not be recognized completely.
3. Recognition difficulties caused by facial changes
At At present, many people have the habit of changing their skin color, such as make-up, in the case of changing skin color, in order to still be able to accurately identify, we need to have the ability to automatically learn the skin color threshold.
4. Face recognition training samples are not enough
It needs a large number of skin color samples to determine the range of skin color. However, due to the cost constraints, the actual number of samples is relatively limited, and the skin color database needs to be expanded through continuous learning. However, this approach may involve stealing user privacy.
5. The false recognition rate is high
On the contrary, when the machine increases the threshold range of skin color, different faces may be recognized by the same face sample. This leads to a decline in the accuracy of face recognition. For practical applications, this problem may be more serious than the low success rate of recognition, which easily leads to personal information security problems.
Post time: Jul-20-2021