Research on finger Extraction

1. combining contrast saliency and region discontinuity for precise hand segmentation in projector-camera system in this article, finger extraction can be considered as three steps. the first is to apply the color histogram to calculate the saliency significance of each pixel of the image and perform an average filter on the Significance itself to obtain an enhanced image, that is, saliency map. Our prospects become more prominent and more suitable for later perform clustering and Other extraction B. then, a clustering algorithm based on the mean shift of the meanshift algorithm is called the meanshift algorithm, which is an iterative process. This method can be applied to the existing code and the results are obtained in L regions. This method is characterized by filtering out the projection pattern. boundary) c. finally, a confidence function is used to obtain the probability that each region belongs to a finger. When the probability is greater than a certain threshold, it is considered to belong to the finger area. Finally, we get the extracted fine finger, as shown in D. So what are the characteristics of this method when it can be applied ???? First, I think saliency is very useful for highlighting the information we want. Of course, the effect is better than simply using the CR channel information in The YCbCr color space. This method highlights the color difference between the foreground and the background, but it has little to do with the color of the foreground, so even black fingers can be used well. Then the meanshift algorithm is a classic clustering method. Compared with methods such as the region Growth algorithm, although iteration is still required, the number of iterations after the saliency map is extracted can be reduced a lot, so there is no need to worry too much about time complexity. The key is that you do not need to extract the centroid. Therefore, you do not need to perform rough extraction on your fingers to find the centroid. This eliminates the issue of incorrect centroid search. Finally, the confidence function method is a good method to determine whether each region belongs to a finger. In the past, simple finger Extraction often extracts the largest Connected Domain directly after marking the connected area, and thinks that this will remove the interference effect. However, sometimes some small connection domains belong to fingers, but they are separated due to various interference. Next, we need to learn more about the implementation details of each step. Reference: 1. global contrast based salient region detection uses significance to divide the image and obtain the details of the highlighted image. mean Shift: a robust approach toward feature space analysis mean offset algorithm for clustering details 3. hand Gesture Recognition in camera-projectorsystem refer to the radiometry calibration section in this article. 2. Global contrast based salient region detection

First, let's raise several questions.

1. What are salient and saliency map?

2. What is the purpose?

3. Under what circumstances?

Salient stands for significance. It generally refers to the obvious and prominent nature of human eyes and their neural vision systems. For example, black spots on the white wall are obvious.

This concept was initially developed from neuroscience. When scientists studied how humans could quickly recognize the visual principles of objects, they found that humans could quickly focus on some areas of the environment, however, they turn a blind eye to other regions. Therefore, this area of attention is called the significance area or saliency region.

Psychologists divide human visual recognition into two situations: top-down and bottom-up.

The top-down recognition mechanism usually occurs in conscious searches for something. We know that we probably know what we want to find, and then perform the significance extraction in front of the slow traversal.

On the contrary, it usually happens in an unconscious situation. We don't know what to look for. It is a fast scanning to locate objective and obvious things. I don't know why. I think the black spot on the wall is so conspicuous.

We know that computer vision algorithms originally exist to simulate human vision. Therefore, based on these two sets of cognitive mechanisms, we developed their respective significance detection methods.

That is, the bottom-up image significance detection and the top-down image significance detection.

The former is data-driven and faster. It is often used to find the significance area in a specific area of the image in terms of color, brightness, direction, and other features. There has been a lot of work in this regard.

The latter is target-driven and is implemented by adjusting the result of the bottom-up method based on the specific task.

There are two cases from top to bottom. Since the target is known, how can this problem be obtained? How do I know the structure and nature of the target?

One is to manually set the detection target. For example, you can perform edge detection by using the canny method. This is achieved only after you have a deep understanding of the structure and nature of the edge (that is, the detection target. But obviously, if our goals are more complex, this human-defined approach is not realistic.

The other is to automatically establish a significance region model through sample training to obtain the nature and structure of the region we are interested in. For example, you can even apply this method to face recognition.

(Similarly, similar methods can be applied for Gesture Recognition)