CN106503611B - Facial image eyeglass detection method based on marginal information projective iteration mirror holder crossbeam - Google Patents
Facial image eyeglass detection method based on marginal information projective iteration mirror holder crossbeam Download PDFInfo
- Publication number
- CN106503611B CN106503611B CN201610814099.8A CN201610814099A CN106503611B CN 106503611 B CN106503611 B CN 106503611B CN 201610814099 A CN201610814099 A CN 201610814099A CN 106503611 B CN106503611 B CN 106503611B
- Authority
- CN
- China
- Prior art keywords
- mirror holder
- region
- holder crossbeam
- pixel
- horizontal line
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/168—Feature extraction; Face representation
- G06V40/171—Local features and components; Facial parts ; Occluding parts, e.g. glasses; Geometrical relationships
Landscapes
- Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- General Health & Medical Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of facial image eyeglass detection methods based on marginal information projective iteration mirror holder crossbeam, mouth region is positioned according to facial image, the direction the x position at mirror holder crossbeam center is determined with mouth region, the direction the y position that mirror holder crossbeam center is determined with the pixel transverse projection situation of the marginal information figure of facial image determines mirror holder crossbeam region according to the direction the x and y position at mirror holder crossbeam center；Horizontal line is searched in mirror holder crossbeam region, if including horizontal line in mirror holder crossbeam region, and horizontal line length is approximately equal to the mirror holder crossbeam region direction x length, then judges that facial image is worn glasses；Otherwise, it is judged as non-wearing spectacles.The present invention can effectively to facial image, whether wearing spectacles be detected, and in detection operation, not only required characteristic information is few, but also can accurately position target area, have obtained a kind of opposite more simple and efficient eyeglass detection method of other methods.
Description
Technical field
The invention belongs to image procossings and computer vision field, and in particular to one kind is based on marginal information projective iteration mirror
The facial image eyeglass detection method of frame crossbeam.
Background technique
In recent years, biometrics identification technology is widely applied to many fields, and face recognition technology is that biological characteristic is known
One of the important topic of other technical research.Currently, face recognition technology has been achieved for great progress, but recognition of face skill
The performance of art nevertheless suffers from the influence of jewelry, posture, expression and other complicated factors.Glasses are that face wears most commonly used decorations
Object, it has been investigated that being influenced with frame glasses on the accuracy rate of human face detection and recognition extremely serious.Glasses detection is in pattern-recognition
It has a wide range of applications with field of image processing, such as, recognition of face, exit and entry certificates shine concrete application according to detection, upload
Piece on-line checking etc..Therefore, whether wearing spectacles have obtained the concern of numerous researchers to detection face in facial image.
Currently, whether the method for wearing spectacles has following a few classes to detection facial image.First kind method is by analyzing eye
Mirror region geometry feature to determine whether wearing spectacles method.Yoshikawa et al. (Glasses frame detection
With 3d hough transform International Conference on Pattern Recognition, IEEE,
2002) in the text by having studied edge feature and geometrical characteristic around glasses, a kind of method for proposing deformable contour outline is come
The presence of glasses is detected, but there are many foundation variable to be considered of this model；(the Automatic eyeglasses such as Wu
Removal from face images, PAMI, 2004) establish facial image of the wearing glasses people not worn glasses corresponding with it
The joint probability distribution model of face image, and the facial image that do not wear glasses is synthesized by this model.This method needs
15 characteristic points are accurately positioned, at this point, the positioning for no-frame glasses is extremely difficult；(the Towards such as Jiang
Detection of glasses in facial images, ICPR, 1998) utilize six spies of glasses surrounding edge information
Sign judges that glasses combine whether there is or not and by this six features and improve detection effect, these features are located in two eyes
Region at the bridge of the nose of the heart and below two eyes, the result that this method positions eyes are very sensitive.Second class method
It is the method based on relative coordinate positioning area-of-interest.Ren Minggang et al. (Glasses detection based on face picture marginal information
Method, software guide, 2014) nasal bridge region positioned according to relative coordinate of the bridge of the nose in facial image, and glasses are carried out accordingly
Detection.This method is avoided that the influence by factors such as illumination, wrinkles in the detection process, but this method is for ridge area
The positioning in domain has great uncertainty in operation.Third class method is the method for statistical learning.Jing et al.
(Glasses detection for face recognition using bayes rules, ICMI, 2000) proposition is based on
The glasses localization method of bayes rule, this method are special according to the glasses that the feature and associative learning of each pixel adjacent domain arrive
Sign is to determine whether the pixel belongs to lens area；Wang et al. (Improvement of face recognition by
Eyeglasses removal, In Proceedings of the Sixth Conference on Intelligent
Information Hiding and Multimedia Signal Processing, 2010) consider glasses and surrounding pixel
Contrast it is obvious, directly extract rims of spectacle using active appearance models AAM method, thus position glasses and extract glasses week
Enclose the characteristic informations such as color, the gradient in region.Such methods based on statistical learning need to carry out feature extraction to great amount of images
And it model is trained can just obtain a result.
Summary of the invention
The object of the present invention is to provide a kind of facial image Glasses detections based on marginal information projective iteration mirror holder crossbeam
Method, solve existing method detect facial image whether wearing spectacles when consider variable it is more, can not accurately position mesh
It marks region and needs the problem of being trained to great amount of images.
The technical scheme adopted by the invention is that the facial image glasses based on marginal information projective iteration mirror holder crossbeam are examined
Survey method is positioned mouth region according to facial image, is determined the direction the x position at mirror holder crossbeam center, with mouth region with face
The pixel transverse projection situation of the marginal information figure of image determines the direction the y position at mirror holder crossbeam center, according to mirror holder crossbeam
The direction the x and y position at center determines mirror holder crossbeam region；Horizontal line is searched in mirror holder crossbeam region, if in mirror holder crossbeam region
Including horizontal line, and horizontal line length is approximately equal to the mirror holder crossbeam region direction x length, then judges that facial image is worn glasses；Otherwise,
It is judged as non-wearing spectacles.
It is above-mentioned that the direction the x position at mirror holder crossbeam center is determined specifically, with the center in the direction mouth region x with mouth region
The direction x position as mirror holder crossbeam center.
The method for positioning mouth region is that original facial image is gone to YCbCr color space, is existed using Gauss model
YCbCr color space calculates the similarity in human face region for mouth region with formula (1), and is switched to bianry image to obtain
Mouth region；
P=exp ((- 0.5) × (x-M) ' inv (cov) × (x-M)) (1)
In formula (1), cov and M respectively indicate the covariance matrix and mean vector of chroma vector；Its value is respectively as follows:
M=(156.5599,117.4361)T (2)
The pixel transverse projection situation of the above-mentioned marginal information figure with facial image determines the direction the y position of mirror holder crossbeam
Specifically, the marginal information figure to facial image carries out transverse projection according to formula (4), the sum of every row pixel R is obtainedj, then press
The position in the direction mirror holder crossbeam y is determined according to formula (5)；
Index (k)=find (Rj> μ × max (R)) (j=1,2,3 ..., m；K=1,2,3 ..., s) (5)；
In formula (4), f ' (i, j) indicates that every pixel value of Image Edge-Detection image f ' (x, y), pro indicate projection fortune
It calculates, RjIndicate the value of the every row of project column vector；In formula (5), R is column vector obtained in formula (4), and max (R) is column vector
Middle maximum value.
Further, μ is threshold coefficient, 0.50≤μ≤0.65.
The method that the above-mentioned direction the x and y position according to mirror holder crossbeam center determines mirror holder crossbeam region is, using formula
(6)-(9) mirror holder crossbeam regional location is calculated,
Gxl=Glassx-(α×W) (6)
Gxr=Glassx+(α×W) (7)
Gyl=Glassy-γ (8)
Gyr=Glassy+(β×H) (9)
In formula (6)-(9), GlassxIt is the center in the mirror holder crossbeam region direction x, GlassyIt is mirror holder crossbeam region y
The position in direction, W are face width, and H is face length, Gxl、GylRespectively cross, the ordinate in the mirror holder crossbeam region upper left corner,
Gxr、GyrRespectively cross, the ordinate in the mirror holder crossbeam region lower right corner；0.010≤α≤0.028,1≤γ≤5,0.03≤β≤
0.08。
The above-mentioned method that whether search horizontal line and interpretation wear glasses in mirror holder crossbeam region are as follows:
(1) in the marginal information in mirror holder crossbeam region, first pixel in every row is searched for line by line from the top down, with
Starting point of first pixel searched as horizontal line, horizontal line length is 1 at this time；
(2) pixel for continuing to test current line and its adjacent uplink and downlink to the right since the horizontal line starting point, if next
Pixel is located at current detection pixel adjacent right side, then judges that it, for a pixel on the horizontal line, adds 1 to horizontal line length,
Otherwise, the horizontal line length computation of current detection row stops, and stops detection；
(3) compare the horizontal line length and the mirror holder crossbeam region direction x length, if the horizontal line length is approximately equal to mirror holder crossbeam
The region direction x length then judges that the image for the image of wearing spectacles, terminates detection；Otherwise, return step (1), according to same
Method continue searching pixel from next line, and calculate the length of next horizontal line, and judged again；If in mirror holder crossbeam
The horizontal line of the mirror holder crossbeam region direction x length is not approximately equal in region, then it represents that the image is the figure of not wearing spectacles
Picture.
It is one of following four situation that step (2) next pixel, which is located at current detection pixel adjacent right side position: under
One pixel is located at the adjacent directly to the right of current detection pixel；Next pixel is located at the adjacent bottom right of current detection pixel
Angle；Next pixel is located at the adjacent upper right corner of current detection pixel；It is just right that next pixel is located at current detection pixel
The position of the one inactive pixels point in side interval.
The horizontal line length is approximately equal to the judgement of the mirror holder crossbeam region direction x length in step (3) specifically: when horizontal line
When length > a × mirror holder crossbeam region the direction x length, it is believed that horizontal line length is approximately equal to the mirror holder crossbeam region direction x length,
0.70≤a≤0.95。
The invention has the advantages that the present invention is to facial image, whether wearing spectacles are detected, with existing detection side
Method is compared, and not only without the concern for multivariable and a large amount of training samples excessively, but also can be simplified to target area precise positioning
Detection operates and improves the performance of detection.
Detailed description of the invention
Fig. 1 is that the present invention is based on the processes of the facial image eyeglass detection method of marginal information projective iteration mirror holder crossbeam
Figure；
Fig. 2 is an image set to be detected of embodiment input；
Fig. 3 is the pre-processed results for the presentation graphics chosen in the image set of Fig. 2；
Fig. 4 is the corresponding edge detection results of image in Fig. 3；
Fig. 5 is the mouth similarity area calculated using Gauss model in YCbCr color space original color image in Fig. 3
The bianry image in domain；
Fig. 6 is the positioning result according to the finally obtained mouth region of Fig. 5, wherein box mark indicates mouth region；
Fig. 7 is according to the center in the mouth region positioning direction x in Fig. 6 come the positioning in the mirror holder crossbeam region direction x determined,
It is indicated with vertical line；
Fig. 8 is the case where carrying out transverse projection to the pixel of Fig. 4 edge detection results image, wherein the curve of no * label
It is the statistical conditions of edge pixel, the curve of * label is after the curve smoothing marked to no * as a result, carrying out subsequent behaviour
When making, with have * mark curve based on；
Fig. 9 is the positioning according to the pixel transverse projection situation of Fig. 8 to the mirror holder crossbeam region direction y, is marked with horizontal line；
Figure 10 is mirror holder crossbeam zone location schematic diagram；
Figure 11 is the crossbeam zone location in Fig. 3 in original color image as a result, being marked with box, and * expression is calculated
The upper left corner in mirror holder crossbeam region and the position in the lower right corner；
Figure 12 is according to interception result of the mirror holder crossbeam region in edge-detected image in Figure 11；
Figure 13 is " potential " horizontal line feature schematic diagram, wherein black region represents working as during searching for " potential " horizontal line
Preceding pixel point, hatched example areas represent next pixel of " potential " horizontal line, and white space indicates inactive pixels point, corresponding edge inspection
Survey the black pixel point in result；
Figure 14 a and Figure 14 b are the horizontal line example detected；* it indicates white pixel point, indicates black pixel point；
Figure 15 a and Figure 15 b are to detect correct result to the facial image of wearing spectacles and non-wearing spectacles respectively.
Specific embodiment
Present invention will be described in further detail below with reference to the accompanying drawings and specific embodiments.
Facial image eyeglass detection method based on marginal information projective iteration mirror holder crossbeam of the invention, first part are
Edge detection is pre-processed and done to facial image, mouth region is positioned with color image, and then determine the side x at mirror holder crossbeam center
To position, the direction the y position of mirror holder crossbeam is determined with the pixel transverse projection situation of edge detection, and with this, it is horizontal to obtain mirror holder
Beam region；Second part searches for " potential " horizontal line in the mirror holder crossbeam region of the marginal information image extracted, according to " latent
" horizontal line judged in the coverage rate of mirror holder crossbeam zone length, with complete to facial image whether the detection of wearing spectacles.
As shown in Figure 1, being specifically implemented according to the following steps:
Step 1, facial image f (x, y) is inputted, as shown in Figure 2.Since there are noises etc. to interfere for collected facial image
Information needs to pre-process face part.Firstly, colorized face images are converted into gray level image；Then, to grayscale image
As carrying out Gaussian smoothing, as shown in Figure 3；Finally, extracting edge f ' (x, y) after gray level image is smoothed, such as
Shown in Fig. 4.
Step 2, the original color image f (x, y) of input is gone into YCbCr color space from RGB color, using height
This model calculated in YCbCr color space with formula (1) be in human face region mouth region similarity P, and switched to two-value
Image obtains possible mouth region, as shown in Figure 5.By the screening of the information such as shape, position, to determine mouth region,
As shown in Figure 6.With the center Glass in the direction mouth region xxAs the direction the x positioning result at mirror holder crossbeam center, such as Fig. 7 institute
Show.
P=exp ((- 0.5) × (x-M) ' × inv (cov) × (x-M)) (1)
Formula (1) is that the colorized face images f (x, y) that will be inputted goes to YCbCr color space from RGB color to carry out
It calculates, wherein x=(Cb, Cr)TFor the chroma vector of pixel；Inverse of a matrix matrix is sought in inv expression；Cov and M distinguishes table
Show the covariance matrix and mean vector of chroma vector；P indicates that the similarity of the pixel and mouth, value are expressed as more greatly mouth
A possibility that portion region, is bigger, otherwise smaller.Through experiment statistics, mean value and variance are respectively as follows:
M=(156.5599,117.4361)T (2)
Step 3, the marginal information figure f ' (x, y) of facial image is subjected in a manner of formula (4) transverse projection, obtains edge letter
The sum of pixel of every row R on breath figure f ' (x, y)j, as shown in Figure 8.The positioning at the mirror holder direction crossbeam y center is determined with formula (5)
As a result Glassy, as shown in Figure 9.
In formula (4), f ' (i, j) indicates that every pixel value of Image Edge-Detection image f ' (x, y), x, y indicate that it is horizontal, vertical
Coordinate, pro indicate project, and the operation that every row element value for realizing to matrix is added obtains a column vector,
RjIndicate the value of the every row of column vector.
Index (k)=find (Rj> μ × max (R)) (j=1,2,3 ..., m；K=1,2,3 ..., s) (5)
In formula (5), R refers to column vector obtained in formula (4)；Max (R) is maximum value in column vector；μ is threshold coefficient, is led to
In normal situation, 0.50≤μ≤0.65 takes 0.55 in the present embodiment；Find refers to be waited when the condition is satisfied, returns to the R for the condition that meetsj
Subscript j；Index (k) records k-th of R for meeting conditionjSubscript, i.e., by the R for the condition that meetsjSubscript j value be assigned to Index
(k)；S records the R for the condition that meetsjNumber；1st meets the R of conditionjLower label be (1), i.e. Index (1) be mirror
The positioning Glass in the direction frame crossbeam yy.The obtained direction mirror holder crossbeam y is as shown in Figure 10.
Step 4, the wide W of face and the high H of face of facial image are obtained.Mirror holder crossbeam region is determined according in step 2 and step 3
Position result (Glassx,Glassy), the region Glass of mirror holder crossbeam is obtained with formula (6)-(9), as shown in figure 11.Existed with the position
Crossbeam region Glass ' is intercepted in marginal information image f ' (x, y), as shown in figure 12.
Gxl=Glassx-(α×W)(6)
Gxr=Glassx+(α×W)(7)
Gyl=Glassy-γ(8)
Gyr=Glassy+(β×H)(9)
In formula (6)-(9), GlassxIt is the center in the mirror holder crossbeam region direction x, GlassyIt is mirror holder crossbeam region y
The position in direction, W are face width, and H is face length, Gxl、GylRespectively cross, the ordinate in the mirror holder crossbeam region upper left corner,
Gxr、GyrRespectively cross, the ordinate in the mirror holder crossbeam region lower right corner.In formula (6), (7), the usual value of α is 0.010≤α
≤ 0.028, in formula (8), (9), the usual value of γ is 1≤γ≤5, and the usual value of β is 0.03≤β≤0.08；In this reality
It applies in example, α=0.025, γ=3, β=0.05.
Similarly, position can also be determined according to the transverse and longitudinal coordinate in the mirror holder crossbeam region upper right corner and the lower left corner.
Step 5, mirror holder crossbeam is modeled as " potential " horizontal line, in the mirror holder crossbeam of the marginal information image extracted
In region, " potential " horizontal line is searched for, if in mirror holder crossbeam region including " potential " horizontal line, and " potential " horizontal line length is somebody's turn to do and is approximately equal to
The mirror holder crossbeam region direction x length, then it represents that mirror holder crossbeam has mirror holder crossbeam in region, then judges that facial image is worn glasses；
Otherwise, it is judged as non-wearing spectacles.
(1) in the marginal information Glass ' in mirror holder crossbeam region, first picture in every row is searched for line by line from the top down
Vegetarian refreshments, using first pixel of the row for searching as the starting point of " potential " horizontal line, J value is the line number of the row, at this time
" potential " horizontal line length is 1；
(2) pixel of current line and its adjacent uplink and downlink is continued to test to the right since " potential " the horizontal line starting point,
As shown in figure 13, the black region in figure is the current pixel point for searching for " potential " horizontal line, and hatched example areas is to search for " potential " cross
Next pixel of line.It is if current detection pixel and next pixel belong to four kinds of positions as shown in fig. 13 that, i.e., next
Pixel is located at the adjacent directly to the right of current detection pixel, the lower right corner, the upper right corner, or is located at current detection pixel directly to the right
When being spaced the position of an inactive pixels point, then it is judged for a pixel on the horizontal line, using formula (10) to horizontal line length
Add 1；
In formula (10), Glass ' (i, J) indicates first pixel of the row, and Glass ' (i, j) expression meets " potential "
Next pixel of horizontal line condition, QL indicate " potential " horizontal line, and the example of " potential " horizontal line QL is as shown in Figure 14 a, 14b, figure
The horizontal line of 14a detection is the horizontal line that the length in the second row pixel is 5, and the horizontal line of Figure 14 b detection is length in the second row pixel
For 2 horizontal line.
If current detection pixel and next pixel are not belonging to four kinds of positional relationships as shown in fig. 13 that, such as next picture
Under vegetarian refreshments and current detection pixel interval are located at current pixel point just there are two above inactive pixels point or next pixel
Situations such as square, is then somebody's turn to do " potential " horizontal line length computation and stops, stops the detection of current line.
The calculating of " potential " horizontal line length is completed in the region according to the method described above.
(3) it is compared with the length of " potential " horizontal line and the mirror holder crossbeam region direction x length, mirror holder crossbeam region x
Direction length is Gxr-Gxl+ 1, give one threshold value a, usual 0.70≤a≤0.95, in the present embodiment, a=0.8.If " latent
" the length len of horizontal lineJ>a×(Gxr-Gxl+ 1), indicate the length of " potential " horizontal line in the mirror holder crossbeam region direction x length
Coverage rate is high, i.e. expression " potential " horizontal line is mirror holder crossbeam, then judges that the image for the image of wearing spectacles, no longer executes subsequent
Judgement.If lenJ≤a×(Gxr-Gxl+ 1), then it represents that the horizontal line may be other situations such as wrinkle or shade, then return step
(1), pixel is continued searching from next line after the same method, judges the cross in next line and its adjacent uplink and downlink region
Line, and the length of next horizontal line is calculated, and judged.If eligible without finding in entire mirror holder crossbeam region
(lenJ>a×(Gxr-Gxl+ 1) horizontal line), then it represents that the image is the image of not wearing spectacles.As shown in figs. 15 a and 15b.
Table 1 gives the accurate of test sample detection of the method for the present invention to the 300 facial images composition acquired on the net
Rate, test sample contain the face under a variety of expressions and a variety of environment, and partial test sample is as shown in Figure 2, wherein wears glasses
Facial image have 177, the facial image that do not wear glasses has 123, the glasses worn include the glasses with frame and
No-frame glasses.
The testing result of the invention of table 1
As can be seen from Table 1, the method for the present invention can accurately detect facial image whether wearing spectacles, work as face
Image can also make also when there is exaggeration expression, dark, face micro- the case where inclining accurate judgement, as Figure 15 a,
Shown in b.Consideration is few in the judgment process by the present invention, it is only necessary to extract the crossbeam region of glasses；Position crossbeam region also simultaneously
Non- is the relative position according to glasses crossbeam in face, but according to mouth region detection and transverse projection from horizontal and vertical
Determine the crossbeam position of glasses；And then wearing spectacles are judged whether according to the marginal information of glasses crossbeam, do not need a large amount of pictures
Training obtain a result.
Above description of the present invention is section Example, and but the invention is not limited in above-mentioned embodiment.
Above-mentioned specific embodiment is schematical, is not restrictive.It is all using method of the invention, do not departing from this hair
In the case of bright objective and scope of the claimed protection, all specific expansions belong within protection scope of the present invention.
Claims (8)
1. the facial image eyeglass detection method based on marginal information projective iteration mirror holder crossbeam, which is characterized in that according to face
Framing mouth region determines the direction the x position at mirror holder crossbeam center, with mouth region with the marginal information figure of facial image
Pixel transverse projection situation determine the direction the y position at mirror holder crossbeam center, according to x the and y direction position at mirror holder crossbeam center
Set determining mirror holder crossbeam region；Horizontal line is searched in mirror holder crossbeam region, if including horizontal line in mirror holder crossbeam region, and horizontal line is long
Degree is approximately equal to the mirror holder crossbeam region direction x length, then judges that facial image is worn glasses；Otherwise, it is judged as non-wearing spectacles；
The method of the positioning mouth region is that original facial image is gone to YCbCr color space, is existed using Gauss model
YCbCr color space calculates the similarity in human face region for mouth region with formula (1), and is switched to bianry image to obtain
Mouth region；
In formula (1), cov and M respectively indicate the covariance matrix and mean vector of chroma vector；X=(Cb, Cr)TFor pixel
Chroma vector；Inverse of a matrix matrix is sought in inv expression；P indicates that the similarity of the pixel and mouth, value are respectively as follows:
M=(156.5599,117.4361)T (2)
2. facial image eyeglass detection method according to claim 1, which is characterized in that described to determine mirror with mouth region
The direction the x position at frame crossbeam center is specifically, using the center in the direction mouth region x as the direction the x position at mirror holder crossbeam center.
3. facial image eyeglass detection method according to claim 1, which is characterized in that the edge with facial image
The pixel transverse projection situation of hum pattern determines the direction the y position of mirror holder crossbeam specifically, marginal information to facial image
Figure carries out transverse projection according to formula (4), obtains the sum of every row pixel Rj, the direction mirror holder crossbeam y is determined according still further to formula (5)
Position；
Index (k)=find (Rj> μ × max (R)) (j=1,2,3..., m；K=1,2,3..., s) (5)；
In formula (4), f ' (i, j) indicates that every pixel value of Image Edge-Detection image f ' (x, y), pro indicate project, Rj
Indicate the value of the every row of project column vector；In formula (5), R be formula (4) obtained in column vector, max (R) be in column vector most
Big value, Index (k) record k-th of R for meeting conditionjSubscript.
4. facial image eyeglass detection method according to claim 3, which is characterized in that in the formula (5), μ is threshold value
Coefficient, 0.50≤μ≤0.65.
5. facial image eyeglass detection method according to claim 1, which is characterized in that described according to mirror holder crossbeam center
The direction the x and y position method that determines mirror holder crossbeam region be to calculate mirror holder crossbeam regional location using formula (6)-(9),
Gxl=Glassx-(α×W) (6)
Gxr=Glassx+(α×W) (7)
Gyl=Glassy-γ (8)
Gyr=Glassy+(β×H) (9)
In formula (6)-(9), GlassxIt is the center in the mirror holder crossbeam region direction x, GlassyIt is the mirror holder crossbeam region direction y
Position, W be face width, H be face length, Gxl、GylRespectively cross, the ordinate in the mirror holder crossbeam region upper left corner, Gxr、
GyrRespectively cross, the ordinate in the mirror holder crossbeam region lower right corner；0.010≤α≤0.028,1≤γ≤5,0.03≤β≤0.08.
6. facial image eyeglass detection method according to claim 1, which is characterized in that described in mirror holder crossbeam region
The method whether search horizontal line and interpretation wear glasses are as follows:
(1) in the marginal information in mirror holder crossbeam region, first pixel in every row is searched for line by line from the top down, with search
Starting point of first pixel arrived as horizontal line, horizontal line length is 1 at this time；
(2) pixel for continuing to test current line and its adjacent uplink and downlink to the right since the horizontal line starting point, if next pixel
Point is located at current detection pixel adjacent right side position, then judges it for a pixel on the horizontal line, add 1 to horizontal line length；
Otherwise, the horizontal line length computation of current detection row stops, and stops detection；
(3) compare the horizontal line length and the mirror holder crossbeam region direction x length, if the horizontal line length is approximately equal to mirror holder crossbeam region
The direction x length then judges that the image for the image of wearing spectacles, terminates detection；Otherwise, return step (1), according to same side
Method continues searching pixel from next line, and calculates the length of next horizontal line, and judged again；If in mirror holder crossbeam region
It is not approximately equal to the horizontal line of the mirror holder crossbeam region direction x length inside, then it represents that the image is the image of not wearing spectacles.
7. facial image eyeglass detection method according to claim 6, which is characterized in that step (2) next pixel
It is one of following four situation that point, which is located at current detection pixel adjacent right side position: next pixel is located at current detection pixel
The adjacent directly to the right of point；Next pixel is located at the adjacent lower right corner of current detection pixel；Next pixel is located at current inspection
Survey the adjacent upper right corner of pixel；Next pixel is located at the position of the one inactive pixels point in current detection pixel directly to the right interval
It sets.
8. facial image eyeglass detection method according to claim 6, which is characterized in that step (3) the horizontal line length
It is approximately equal to the judgement of the mirror holder crossbeam region direction x length specifically: when length > a × mirror holder crossbeam region direction x of horizontal line is long
When spending, it is believed that horizontal line length is approximately equal to the mirror holder crossbeam region direction x length；0.70≤a≤0.95.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610814099.8A CN106503611B (en) | 2016-09-09 | 2016-09-09 | Facial image eyeglass detection method based on marginal information projective iteration mirror holder crossbeam |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201610814099.8A CN106503611B (en) | 2016-09-09 | 2016-09-09 | Facial image eyeglass detection method based on marginal information projective iteration mirror holder crossbeam |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106503611A CN106503611A (en) | 2017-03-15 |
| CN106503611B true CN106503611B (en) | 2019-11-22 |
Family
ID=58291395
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201610814099.8A Active CN106503611B (en) | 2016-09-09 | 2016-09-09 | Facial image eyeglass detection method based on marginal information projective iteration mirror holder crossbeam |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106503611B (en) |
Families Citing this family (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3699808B1 (en) | 2017-11-14 | 2023-10-25 | Huawei Technologies Co., Ltd. | Facial image detection method and terminal device |
| CN107945126B (en) * | 2017-11-20 | 2022-02-18 | 杭州登虹科技有限公司 | Method, device and medium for eliminating spectacle frame in image |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103093210A (en) * | 2013-01-24 | 2013-05-08 | 北京天诚盛业科技有限公司 | Method and device for glasses identification in face identification |
| CN104408426A (en) * | 2014-11-27 | 2015-03-11 | 小米科技有限责任公司 | Method and device for removing glasses in face image |
| CN105787427A (en) * | 2016-01-08 | 2016-07-20 | 上海交通大学 | Lip area positioning method |
-
2016
- 2016-09-09 CN CN201610814099.8A patent/CN106503611B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103093210A (en) * | 2013-01-24 | 2013-05-08 | 北京天诚盛业科技有限公司 | Method and device for glasses identification in face identification |
| CN104408426A (en) * | 2014-11-27 | 2015-03-11 | 小米科技有限责任公司 | Method and device for removing glasses in face image |
| CN105787427A (en) * | 2016-01-08 | 2016-07-20 | 上海交通大学 | Lip area positioning method |
Non-Patent Citations (2)
| Title |
|---|
| 一种新型的带反馈的人眼检测方法;张长国;《可编程控制器与工厂自动化》;20120415(第4期);第73页第4节 * |
| 基于人脸图片边缘信息的眼镜检测;任明罡 等;《软件导刊》;20140731;第13卷(第7期);第142-143页第1.3-1.4节，图4-5 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106503611A (en) | 2017-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Lin | Face detection in complicated backgrounds and different illumination conditions by using YCbCr color space and neural network | |
| CN106709568B (en) | The object detection and semantic segmentation method of RGB-D image based on deep layer convolutional network | |
| CN105205480B (en) | Human-eye positioning method and system in a kind of complex scene | |
| CN105844252B (en) | A kind of fatigue detection method of face key position | |
| CN100423020C (en) | Human face identifying method based on structural principal element analysis | |
| CN110348319A (en) | A kind of face method for anti-counterfeit merged based on face depth information and edge image | |
| CN108898125A (en) | One kind being based on embedded human face identification and management system | |
| CN102324025A (en) | Human face detection and tracking method based on Gaussian skin color model and feature analysis | |
| CN107368778A (en) | Method for catching, device and the storage device of human face expression | |
| CN106022231A (en) | Multi-feature-fusion-based technical method for rapid detection of pedestrian | |
| CN106503644B (en) | Glasses attribute detection method based on edge projection and color characteristic | |
| CN104794693B (en) | A kind of portrait optimization method of face key area automatic detection masking-out | |
| CN106570447B (en) | Based on the matched human face photo sunglasses automatic removal method of grey level histogram | |
| CN106599785A (en) | Method and device for building human body 3D feature identity information database | |
| CN114359998B (en) | Identification method of face mask in wearing state | |
| CN106611158A (en) | Method and equipment for obtaining human body 3D characteristic information | |
| CN110533648A (en) | A kind of blackhead identifying processing method and system | |
| CN106909884A (en) | A kind of hand region detection method and device based on hierarchy and deformable part sub-model | |
| CN110929570B (en) | Iris rapid positioning device and positioning method thereof | |
| CN106503611B (en) | Facial image eyeglass detection method based on marginal information projective iteration mirror holder crossbeam | |
| Montazeri et al. | Automatic extraction of eye field from a gray intensity image using intensity filtering and hybrid projection function | |
| Yi et al. | Face detection method based on skin color segmentation and facial component localization | |
| Patil et al. | Automatic detection of facial feature points in image sequences | |
| CN112183215A (en) | Human eye positioning method and system combining multi-feature cascade SVM and human eye template | |
| Al-Hameed et al. | Face detection based on skin color segmentation and eyes detection in the human face |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20210325 Address after: 1213-156, building 3, No. 1366, Hongfeng Road, Kangshan street, Huzhou Economic and Technological Development Zone, Zhejiang Province, 313000 Patentee after: Zhejiang Zhongchuang Yuhao Information Technology Co.,Ltd. Address before: 710048 No. 5 Jinhua South Road, Shaanxi, Xi'an Patentee before: XI'AN University OF TECHNOLOGY |
|
| TR01 | Transfer of patent right |