[0040] Next, the technical solutions in the embodiments of the present invention will be described in contemplation in the embodiment of the present invention, and it is clear that the described embodiments are intended to be in an embodiment of the invention, not all of the embodiments of the invention. Based on the embodiments of the present invention, there are all other embodiments obtained without making creative labor without making creative labor premises.
[0041] The present invention provides a human face image feature point assistant label method, specifically divided into three parts: main feature point pre-label, secondary feature point assist label, manual adjustment feature point, specific implementation process figure 1 Indicated. In this program, a primary feature point design can be combined with a variety of secondary feature points, for 22 main feature points, such as image 3 As shown, after combination with four common secondary feature points, as shown in Table 1, 44, 68, 72, 106 feature points are obtained. That is, a variety of feature points can be achieved by training a model for detecting 22 points. Figures 4 (a) -4 (d) They are corresponding to 22, 46, 50, 64 secondary feature points.
[0042] Table 1 Distribution of Face Feature Point
[0043]
[0044] 1. Main feature point is expected
[0045] The main fastening point is expected to be implemented by the pre-training model. Currently, algorithms for training feature point detection models include ASM (Active of Regression Trees), based on deep learning. In order to improve the accuracy of the main feature point label, this program uses a deep learning method training model. After the model is detected, 22 primary feature points are detected, and the pre-scaled note is checked by the user. If the specified indicator is not reached, the user is corrected to ensure the accuracy of the main feature point label. The accuracy of strengthening the main feature point is conducive to increasing the next step characteristic point assistance positioning accuracy.
[0046] 2. Secondary feature point auxiliary label
[0047]The position of the secondary feature point is determined by the equivalent distance distribution method based on the edge information of the two main feature points and the face of the face. In order to better detect the edge of the human face and calculate the distribution of the secondary feature point, this program divides the main feature points according to the facial features of the face to 5 parts: eyebrows, eyes, nose, mouth, face, and then The secondary feature point of each part is aided, and finally combines the secondary feature point to facilitate manual adjustment operations in the next step.
[0048] The role of edge detection is to detect the contours of the faces and as the positioning basis for the secondary feature point. There are many existing edge detection algorithms, such as differential edge detection, Sobel operator, canny operator, etc., where the Canny operator is an edge detection method employed in this scheme. The Canny operator has strong anti-noise capacity, accurate positioning and single edge detection advantage, making it more suitable for detecting complete face edge information. In this scenario, in addition to the image segmentation, the edge detection is reduced, and the edge detection result is also required to perform morphological processing and edge subsacking treatment. Generally, the image detected by the edge will have a large number of interference noise, and the morphological processing of the image can effectively filter image noise. However, after forming form, the edges have fracture, but also connect the main feature points to the remaining edge to form a closed loop and used to calculate the sequential point position.
[0049] The secondary feature point position is distributed on the equipped two main feature points, such as figure 2 As shown, A, B indicates the main feature point, C, D represent the secondary feature point, and e represents the edge line. Where A coordinates are (X 1 Y 1 ), The coordinates of B is (X 2 Y 2 ), The AB connection and horizontal line angle is θ:
[0050]
[0051] C The intersection of the equidistant line and the AB connection is
[0052] D The intersection of the equidistant line and the AB connection is
[0053] Set C coordinates (x c Y c ), The equidistor equation of C is:
[0054]
[0055] Set D's coordinates (X d Y d ), The equidistor equation of D is:
[0056]
[0057] Therefore, the coordinates of the feature points C, D can be achieved according to the intersection of the equidistant line and the E of C, D. The auxiliary label function can be realized. The increase in secondary feature points or reduces the auxiliary labeling function of different tasks only by increasing or reducing the equidist lines between AB. Further, the equidistant line equation can be confined to the position of the second feature. When the manual adjustment position is required, the secondary feature point can only be moved on the isometric line, so that the label deviation produced by the human factors can be reduced.
[0058] In terms of existing programs, the labeling efficiency is not high, and the applicability is highly affected by the objective factors. The present invention splits the human face feature point into two types of primary feature points, wherein the primary characteristic point is expected, the secondary feature point assist label, and then manually determines whether or not to adjust the feature point position. In a face image, the main feature point is composed of positions that are relatively fixed and can indicate the key information of the face's key information. The 22 main feature points labeled in this program image 3 Indicated. The secondary feature point is determined by the adjacent two main feature points and human face edge information, and its position can be flexible according to demand. This program makes full use of the primary feature point pre-standard and secondary feature point assist labeling method to increase the labeling efficiency; the main feature point is combined with different feature points to enhance the applicability of the labeling system; the equivalence Distributed constraints to alleviate the margin deviation caused by subjective factors.
[0059] In summary, this program first divides the feature point into the primary feature point and the secondary feature point, and the use of the primary feature point and the secondary feature point combination of different labels can achieve a variety of labeling schemes. purpose. Second, the feature points are divided into 5 parts according to the characteristics of the characteristics, which improves the quality of the edge extraction, and it is also convenient for the secondary feature point distribution position. Then use the geometric relationship between the adjacent primary feature points and the edge line information to estimate the position of the feature point. Finally, the secondary feature point is constrained by equal distance distribution, making the secondary feature point position label more standardized.
[0060] This scheme has a variety of design in the primary segmentation point, and a primary feature point can be combined with a variety of secondary feature points, while the primary feature point can adjust the quantity and distribution position according to the shape and edge information of the target and the shape of the object. In addition, the feature point labeling technique of this program has certain versatility, in addition to the application of the face feature point label, can also be applied to other characteristic points labeling scenes with clear edges, such as pedestrians, feature points of the vehicle.
[0061] The specific embodiments described above are described in further detail purposes of the present invention, and it is understood that only the embodiments of the present invention are not intended to limit the invention. protected range. Any modifications, equivalent replacement, improvement, etc., without departing from the spirit and scope of the present invention, is also within the scope of the invention.
