WO2017092589A1 - 一种图像中人像轮廓的确定方法和装置 - Google Patents

一种图像中人像轮廓的确定方法和装置 Download PDF

Info

Publication number
WO2017092589A1
WO2017092589A1 PCT/CN2016/106850 CN2016106850W WO2017092589A1 WO 2017092589 A1 WO2017092589 A1 WO 2017092589A1 CN 2016106850 W CN2016106850 W CN 2016106850W WO 2017092589 A1 WO2017092589 A1 WO 2017092589A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
area
detection window
unit
color
Prior art date
Application number
PCT/CN2016/106850
Other languages
English (en)
French (fr)
Inventor
李川石
Original Assignee
阿里巴巴集团控股有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 阿里巴巴集团控股有限公司 filed Critical 阿里巴巴集团控股有限公司
Publication of WO2017092589A1 publication Critical patent/WO2017092589A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/40—Analysis of texture
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10024—Color image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20112—Image segmentation details
    • G06T2207/20116—Active contour; Active surface; Snakes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30196—Human being; Person

Definitions

  • the present invention relates to the field of image processing, and in particular to a method and apparatus for determining a silhouette of a portrait in an image.
  • the objects here can mainly include buildings, portraits, and the like.
  • the determined object contour can be used for later image processing, processing, and the like.
  • these contour determination methods for the contour of an object mainly pursue the speed of contour determination, that is, it is desirable to recognize the contour of the object as soon as possible.
  • the attention to the accuracy of object contour recognition is not high.
  • the recognition accuracy of the silhouette of the portrait is required to be high.
  • the contour of the conventional object is used to process the obtained silhouette of the figure, and the accuracy of the silhouette of the figure is difficult to meet the requirements for the accuracy of the silhouette of the portrait in subsequent processing.
  • the present invention provides a method and apparatus for determining a silhouette of a portrait in an image, and a more accurate determination result of the contour of the portrait is obtained by moving the detection window.
  • a method for determining a silhouette of a portrait in an image comprising:
  • the length of the detection window being a plurality of unit lengths, the number of length units being predetermined according to the image, the detection window including a unit length Color detection area;
  • the color value of the area passing by the color detection area on the image is recorded; and during the moving of the detection window, the first average value is continuously updated, the An average value is the average value of the color of the recorded color value per unit length;
  • the difference between the color value of the first region and the first average value exceeds the first preset range stopping the moving the detection window, and determining the first average value and Difference between the second average Whether the value exceeds a second preset range, wherein the first area is an area in the image that is covered by the color detection area, and the second average value is a color value of the first area and the first a color average value of the second area at a unit length when the difference between the average values exceeds the first predetermined range, and the second area is an area in the image that is covered by the detection window;
  • the second preset range determines that the difference between the color value of each unit length in the second area and the first average value exceeds the number of the second preset range, if the number is exceeded.
  • the ratio between the number of unit lengths of the two preset ranges and the total number of unit lengths of the second area does not exceed a preset ratio, and the first area is determined to be a part of the silhouette of the portrait in the image.
  • the method further includes:
  • the color detection area is located at a last position in the detection window along a horizontal movement direction.
  • the determining whether the difference between the first average value and the second average value exceeds a second preset range further includes:
  • the second preset range is not exceeded, continue to move the detection window, and determine that the first area is a pixel impurity in the image, and the color value of the first area is no longer used to determine the The position of the portrait outline in the image.
  • the determining that the difference between the color value of each unit length in the second area and the first average value exceeds the number of the second preset range further includes:
  • the ratio between the number of unit lengths exceeding the second preset range and the total number of unit lengths of the second area exceeds a preset ratio, continue to move the detection window, and determine the first area
  • the color values of the first region are no longer used to determine the location of the silhouette of the portrait in the image.
  • the method before the horizontally moving the preset detection window from one side of the image to the other side, the method further includes:
  • the color model of the image is converted to a hue saturation luminance HSV color model.
  • the method before the horizontally moving the preset detection window from one side of the image to the other side, the method further includes:
  • the length of the detection window is a plurality of unit lengths, and the number of the length units is determined according to the image, and specifically includes:
  • the number of unit lengths of the length of the detection window is determined according to the shape parameter of the rectangular area.
  • a device for determining a silhouette of a portrait in an image comprising:
  • a moving unit configured to horizontally move a preset detection window from one side of the image to the other side, the length of the detection window being a plurality of unit lengths, the number of length units being predetermined according to the image, the detecting
  • the window includes a color detection area of unit length;
  • a recording unit configured to record, in a process of moving the detection window, a color value of an area on the image that is passed by the color detection area; and continuously update the first average during the moving of the detection window a value, the first average value being a color average value of the recorded color value per unit length;
  • a first determining unit configured to stop moving the detection window if a difference between a color value of the first region and the first average value exceeds a first preset range during moving the detection window, And determining whether the difference between the first average value and the second average value exceeds a second preset range, wherein the first area is an area in the image that is covered by the color detection area, and second The average value is a color average value of the second region at a unit length when the difference between the color value of the first region and the first average value exceeds the first predetermined range, and the second region is the An area in the image that is covered by the detection window; if the second preset range is exceeded, the second determining unit is triggered,
  • the second determining unit is configured to determine that the difference between the color value of each unit length in the second area and the first average value exceeds the number of the second preset range, if the second part is exceeded.
  • the ratio between the number of unit lengths of the preset range and the total number of unit lengths of the second area does not exceed a preset ratio, and the first area is determined to be a part of the silhouette of the portrait in the image.
  • the moving unit is further configured to: after triggering the second determining unit, shift the detection window to a width perpendicular to the horizontal moving direction, and horizontally move from one side of the image to the other side
  • the detection window continues to determine the position of the silhouette of the portrait in the image until the image is traversed through the detection window.
  • the color detection area is located at a last position in the detection window along a horizontal movement direction.
  • it also includes:
  • the determining unit is configured to determine that the first area is a pixel impurity in the image, and the color value of the first area is no longer used to determine a position of a silhouette of the portrait in the image.
  • it also includes:
  • the determination result of the second determining unit is that the ratio between the number of unit lengths exceeding the second preset range and the total number of unit lengths of the second area exceeds a preset ratio, triggering the movement Unit continues to move The window is detected and the determining unit is triggered.
  • it also includes:
  • a converting unit configured to convert the color model of the image into a hue saturation brightness HSV color model before triggering the moving unit.
  • it also includes:
  • a matching unit configured to determine, by using face matching, a rectangular area in which the face of the portrait in the image is located before triggering the mobile unit;
  • the length of the detection window is a plurality of unit lengths, and the number of the length units is determined according to the image, and specifically includes:
  • the number of unit lengths of the length of the detection window is determined according to the shape parameter of the rectangular area.
  • the image is recorded on the image by the color detection area. a color value of the region; and during the moving of the detection window, continuously updating the first average value, in the process of moving the detection window, if a color value of the first region is between the first average value and If the difference exceeds the first preset range, stop moving the detection window, and determine whether the difference between the first average value and the second average value exceeds a second preset range, if the second preset range is exceeded, And the ratio between the number of unit lengths exceeding the second preset range and the total number of unit lengths of the second area does not exceed a preset ratio, and determining that the first area is a silhouette of a portrait in the image portion.
  • the detection is performed by moving the monitoring window.
  • the color value can be collected by moving horizontally from one side of the image to the other side, so that the images can be more accurately and effectively analyzed.
  • the change in the color value of the line results in a more accurate determination of the silhouette of the portrait.
  • FIG. 1 is a flowchart of a method for determining a silhouette of a portrait in an image according to an embodiment of the present invention
  • FIG. 2 is a schematic diagram of a face detection area in an image according to an embodiment of the present invention.
  • FIG. 3 is a schematic diagram of determining a silhouette of a portrait in an image according to an embodiment of the present invention.
  • FIG. 4 is a schematic diagram of determining a silhouette of a portrait according to an embodiment of the present invention.
  • FIG. 5 is a structural diagram of an apparatus for determining a silhouette of a portrait in an image according to an embodiment of the present invention.
  • the recognition accuracy of the silhouette of the portrait is required to be high.
  • the contour determination method of the conventional object contour determination method is not high in accuracy, and the edge processing of the silhouette of the portrait is unclear, or a part of the hair or the effective part of the portrait is cropped. It can be seen that the accuracy of the traditional contour determination method is difficult to meet the requirements of post processing.
  • a method and apparatus for determining a silhouette of a portrait in an image provided by an embodiment of the present invention, by moving a preset detection window horizontally from one side of the image to the other side, in the process of moving the detection window, recording a color value of an area on the image that is passed by the color detection area; and in a process of moving the detection window, continuously updating a first average value, in the process of moving the detection window, if the first area
  • the difference between the color value and the first average value exceeds the first preset range, stopping moving the detection window, and determining whether the difference between the first average value and the second average value exceeds the second pre-predetermined a range, if the second preset range is exceeded, and the ratio between the number of unit lengths exceeding the second preset range and the total number of unit lengths of the second area does not exceed a preset ratio, determining
  • the first area is part of a silhouette of a portrait in the image.
  • the detection is performed by moving the monitoring window.
  • the data to be collected and processed is relatively large, the data processing time is long, but the method of collecting the color value by moving horizontally from one side of the image to the other side can be more accurate. Effectively analyzing changes in the color values of the images in the same row, thereby obtaining a more accurate determination of the silhouette of the portrait.
  • the executor of the present invention may be a computer program that can be installed and deployed at the end of the mobile End, computer, server, etc.
  • the mobile terminal, the computer, the server, and the like can implement the image processing flow of the image provided by the embodiment of the present invention by running the computer program deployed in the embodiment to determine the outline of the portrait in the image.
  • the detection window appearing below can be understood as the detection interval of the color value of the program on the image.
  • the program determines the detection interval of the color value (that is, the detection window) after determining the edge of the image, and detects the color value of the image portion in the detection interval falling within the color value.
  • the detection interval of the color value is moved on the image as described below, and the color value of the image portion falling within the detection interval of the color value is continuously detected.
  • the color detection zone appearing below can be understood as a specific, fixed position in the detection window, or a specific area in the detection interval of this color value. During the movement of the detection interval of this color value, the position of the color detection area in the detection interval of this color value is relatively fixed.
  • the program may save the detected color value, the calculated average value, and the like in a storage medium such as a memory or a cache, or in a specially-defined storage area, so that the image processing requires these parameters. transfer.
  • FIG. 1 is a flowchart of a method for determining a silhouette of a portrait in an image according to an embodiment of the present invention, where the method includes:
  • S101 horizontally shifting a preset detection window from one side of the image to the other side, wherein the length of the detection window is a plurality of unit lengths, and the number of length units is predetermined according to the image, and the detection window includes one Color detection area per unit length.
  • an image described in the embodiment of the present invention can be generally understood as a portrait image having a light background.
  • the light colors described herein may be common background colors such as white, light gray, and the like.
  • Portrait images can be used for common half-length documents.
  • the color model of the image may be adjusted, and optionally, the color model of the image may be converted.
  • Color model for hue saturation English: Hue, Saturation, Value, abbreviation: HSV). This color model is more accurate than the common color model used in the image, such as the red, green, blue (abbreviation: RGB) color model, and the information contained is more accurate. Conducive to subsequent image processing.
  • the shape of the detection window is related to the pixel shape, and specifically may be a rectangular shape.
  • the length of the detection window is a plurality of unit lengths, and the number of length units thereof is predetermined according to the image. It should be noted that, according to a specific image situation, the side of the slave image is horizontal to the other side.
  • Embodiment of the present invention before moving a preset detection window At least two ways of setting the length of the detection window are provided. However, it should be noted that the length of the detection window needs to be at least two unit lengths regardless of the manner in which the length of the detection window is set.
  • a unit length is equal to a few pixels, which can be determined according to different application scenarios or portrait contour determination accuracy.
  • a unit length may be equal to an integer number of pixels, for example, may be equal to one pixel or equal to two pixels.
  • the position and size of the portrait in the general photo can conform to a certain size ratio, so for this type of photo, the detection window
  • the length can be set according to the size of the photo (that is, the image).
  • the second way can be applied to different types of images with portraits, that is, the length of the detection window is set by the shape parameters (for example, width, length) of the rectangular area in which the face is located. That is, optional, through the face matching, the rectangular area where the face of the portrait in the image is located is determined.
  • the length of the detection window in S101 is a plurality of unit lengths, and the number of the length units is determined according to the image, and specifically includes: determining, according to the shape parameter of the rectangular area, a unit length of the length of the detection window. number.
  • Face matching can be performed through an image processing framework, which is not limited by the present invention. A rectangular area or a face detection area is generally determined.
  • FIG. 2 is a schematic diagram of a face detection area in an image according to an embodiment of the present invention.
  • W is the width of the face detection area
  • H is the length of the face detection area
  • 10 is the detection window. Shown in Fig. 2 is the case where the detection window is moved from the left side to the right side of the image.
  • the embodiment of the present invention further provides a specific calculation method for determining the number of unit lengths of the detection window according to the shape parameter of the rectangular region, and the detection is performed on the premise that one unit length is equal to one pixel.
  • the length of the window can be half the width of the portrait detection area.
  • the present invention does not limit the position of the color detection area in the detection window. To ensure the accuracy of the portrait contour determination, only the length of the color detection area is limited to one unit length. In the movement mode and the calculation mode provided by the embodiment of the present invention, generally, the tail portion located in the moving direction can achieve the best figure contour determination accuracy. That is to say, optionally, the color detection zone may be located at the last position in the detection window along the horizontal movement direction. 11 as shown in FIG. 2 may be the position of the color detection area in the detection window, that is, the leftmost side, when moving horizontally from the left side to the right side of the image. If the detection window is horizontally moved from the right side to the left side of the image, the color detection area is at the rightmost side of the detection window.
  • the manner of horizontal movement from one side of the image to the other side is generally limited to moving from the left side to the right side of the image or from the right side to the left side, but there is no embodiment in the present invention.
  • the placement of the image is defined, the image being placed as shown in Figure 2, i.e., in a normal orientation when photographed (generally the portrait of the portrait is facing up).
  • the image can also be placed sideways, for example, the placement angle shown in FIG. 2 is rotated clockwise or counterclockwise by 90 degrees, in the case of such side placement, from one side of the image to the other
  • the manner in which the side moves horizontally can be understood as moving horizontally from below the image to above, or horizontally from above the image to below.
  • S102 during the moving of the detection window, record a color value of an area on the image that is passed by the color detection area; and continuously update the first average value during the process of moving the detection window,
  • the first average value is the average value of the color of the recorded color value per unit length.
  • FIG. 3 is a schematic diagram of determining a silhouette of a portrait in an image according to an embodiment of the present invention.
  • an area passing by the color detection area on the image may be the color.
  • the sum of the areas covered by the detection area 11 in the course of the movement is the same.
  • the size of the area divided by each of the dotted squares is the same as the size of the color detection area 11, and the length is one unit length.
  • the area on the image that is passed by the color detection area includes the dotted line area shown in FIG. 3 and the area that the color detection area is currently covering.
  • the recorded color value may be a color value of each pixel in the region through which the color detection region passes, and when the first average value is calculated, not calculated in units of pixels, but in unit length Unit calculation. Assuming that the area passed by the color detection area on the image has 100 pixels and one unit length is two pixels, then when the first average value is calculated in step S102, the color values of 100 pixels are summed. Divide by 50 (ie 100 pixels for 50 unit lengths).
  • S103 in the process of moving the detection window, if the difference between the color value of the first area and the first average value exceeds the first preset range, stop moving the detection window, and determine the first average Whether the difference between the value and the second average value exceeds a second predetermined range, wherein the first area is an area in the image that is covered by the color detection area, and the second average is a color average value of the second area at a unit length when the difference between the color value of the area and the first average value exceeds the first predetermined range, and the second area is the image in the image The area in which the detection window covers the state.
  • the first area is an area in the image covered by the color detection area, and the first area moves as the detection window moves on the image.
  • the second area also moves as the detection window moves over the image.
  • the specific location can be seen in Figure 3.
  • the second average value is when the detection window stops, because the detection window stops when the difference between the color value of the first region and the first average value exceeds the first preset range. Covered image area The average of the color values in the middle. When calculating the second average value, it is still necessary to note that it is not calculated in units of pixels, but is calculated in units of unit length.
  • the first preset range and the second preset range may be the same or different, and may be, for example, ⁇ 20%.
  • it can be used to determine a sudden change in the color value of the image on the moving path of the detection window, and the point of the mutation is in the first area.
  • this mutation may be an impurity in the image background of the image, or may be the position of the silhouette of the portrait in the image, which is specifically possible, and can be effectively distinguished and identified by the judgment in S104. .
  • S104 If the second preset range is exceeded, determining that the difference between the color value of each unit length in the second area and the first average value exceeds the number of the second preset range, if The ratio between the number of unit lengths of the second preset range and the total number of unit lengths of the second area does not exceed a preset ratio, and the first area is determined to be a part of the silhouette of the portrait in the image.
  • the trend of the color value change in the second region may be determined. Whether it is smooth or not, thereby judging whether the detection window is currently in the area of the portrait in the image.
  • the preset ratio may be set according to a specific application scenario or determining the accuracy requirement of the silhouette of the portrait, and generally set a smaller value, for example, 25%.
  • FIG. 4 is a schematic diagram of determining a silhouette of a portrait according to an embodiment of the present invention. It can be seen that when the detection window enters the portrait area and the color detection area is located on the portrait contour, this can be identified by the judgment in S104.
  • the position of the silhouette of the portrait on the side of the moving direction of the detection window at the moving height of the image movement can be determined. Thereafter, by changing the moving height and the moving direction of the detection window, the detected train is used to traverse the image, and the schemes of S101 to S104 are repeated during the traversal, thereby determining the complete portrait outline of the portrait in the image.
  • the method further includes:
  • the width of the detection window is described.
  • the detection window in order to improve the detection of the contour of the portrait as much as possible
  • it is desirable to set the detection window as narrow as possible that is to say, the narrower the detection window, the higher the accuracy of determining the silhouette of the portrait of the portrait in the image using the detection window.
  • An embodiment of the present invention provides an optional width of the detection window, that is, one pixel.
  • the width of the detection window can be adaptively adjusted according to the detection accuracy requirements of the specific scene.
  • the detection window After determining the width of the detection window, when the detection window determines the position of a portrait contour, the detection window is returned to one side of the image, and the detection is moved up or down at the moving height. The width of the window continues to move the detection window to determine the position of the portrait of the portrait in the image at this moving height. In the process of traversing the image, it may be moved from one side to the other side, and then traversed in the reverse direction, for example, the detection window may be moved from the left edge to the right edge from the upper edge of the image, and the completion is completed.
  • the moving height of the detection window is moved downward, and after traversing the image from the top to the bottom, the positions of the determined portrait contours are stitched together to form a contour on the left side of the portrait in the image. Then, starting from the uppermost edge of the right side of the image, traversing the image from top to bottom in the direction from the right side to the left side of the image, after the traversal is completed, the determined silhouette of the portrait The positions are stitched together to form a contour to the right of the portrait in the image, and the previously determined left contour is combined to form a complete portrait of the portrait in the image.
  • the image is recorded on the image by the color detection area. a color value of the region; and during the moving of the detection window, continuously updating the first average value, in the process of moving the detection window, if a color value of the first region is between the first average value and If the difference exceeds the first preset range, stop moving the detection window, and determine whether the difference between the first average value and the second average value exceeds a second preset range, if the second preset range is exceeded, And the ratio between the number of unit lengths exceeding the second preset range and the total number of unit lengths of the second area does not exceed a preset ratio, and determining that the first area is a silhouette of a portrait in the image portion.
  • the detection is performed by moving the monitoring window.
  • the color value can be collected by moving horizontally from one side of the image to the other side, so that the images can be more accurately and effectively analyzed.
  • the change in the color value of the line results in a more accurate determination of the silhouette of the portrait.
  • the determining whether the difference between the first average value and the second average value exceeds a second preset range further includes:
  • the second preset range is not exceeded, continue to move the detection window, and determine that the first area is a pixel impurity in the image, and the color value of the first area is no longer used to determine the The position of the portrait outline in the image.
  • the difference between the first average value and the second average value does not exceed the second preset range, it may be determined that the color value change in the second area is described on the image.
  • the difference between the color value changes of the area through which the color detection area passes is not large, then the detection window may not completely enter the area of the portrait in the image, and then the position covered by the color detection area in the detection window It cannot be used as a portrait outline of a portrait in the image. It is therefore possible to determine that the first region is an impurity in the background of the portrait on the image, or a point that is unacceptable in the image.
  • Such impurity points or unacceptable points can affect the accuracy of the subsequent contour determination of the portrait, so to ensure accuracy, the color values of the first region are no longer used to determine the position of the silhouette of the portrait in the image. For example, this color value or the like can be deleted from the color value recorded in S102.
  • the determining that the difference between the color value of each unit length in the second area and the first average value exceeds the number of the second preset range further includes:
  • the ratio between the number of unit lengths exceeding the second preset range and the total number of unit lengths of the second area exceeds a preset ratio, continue to move the detection window, and determine the first area
  • the color values of the first region are no longer used to determine the location of the silhouette of the portrait in the image.
  • the color value change in the second area may be determined.
  • the trend is not smooth, then the detection window may not fully enter the area of the portrait in the image, then the position covered by the color detection area in the detection window is not a portrait of the portrait in the image. profile.
  • the difference between the first average value and the second average value has exceeded the second predetermined range, the difference between the color value of the first region and the first average value has also exceeded the first pre-predetermined The range is set, so it can be determined that the color value of the pixel point in the first area has suddenly changed.
  • the first region is an impurity in the background of the portrait on the image, or a point that is unacceptable in the image.
  • impurity points or unacceptable points can affect the accuracy of the subsequent contour determination of the portrait, so to ensure accuracy, the color values of the first region are no longer used to determine the position of the silhouette of the portrait in the image. For example, this color value or the like can be deleted from the color value recorded in S102.
  • the process of identifying the pixel outline or the unacceptable point in the image and no longer using the color value of the impurity or the unacceptable point for determining the position of the silhouette of the portrait in the image Therefore, the accuracy of determining the position of the silhouette of the portrait can be effectively improved.
  • FIG. 5 is a structural diagram of an apparatus for determining a silhouette of a portrait in an image according to an embodiment of the present invention, where the apparatus includes:
  • a moving unit 501 configured to horizontally move a preset detection window from one side of the image to the other side, wherein the length of the detection window is a plurality of unit lengths, and the number of length units is predetermined according to the image,
  • the detection window includes a color detection area of unit length.
  • it also includes:
  • a converting unit configured to convert the color model of the image into a hue saturation brightness HSV color model before triggering the moving unit.
  • it also includes:
  • a matching unit configured to determine, by using face matching, a rectangular area in which the face of the portrait in the image is located before triggering the mobile unit;
  • the length of the detection window is a plurality of unit lengths, and the number of the length units is determined according to the image, and specifically includes:
  • the number of unit lengths of the length of the detection window is determined according to the shape parameter of the rectangular area.
  • the color detection area is located at a last position in the detection window along a horizontal movement direction.
  • a recording unit 502 configured to record, in a process of moving the detection window, a color value of an area on the image that is passed by the color detection area; and continuously update the first in the process of moving the detection window
  • the average value is the average value of the color of the recorded color value per unit length.
  • a first determining unit 503 configured to stop moving the detection window if a difference between a color value of the first region and the first average value exceeds a first preset range during moving the detection window And determining whether the difference between the first average value and the second average value exceeds a second predetermined range, wherein the first area is an area in the image that is covered by the color detection area, The second average value is a color average value of the second area at the unit length when the difference between the color value of the first area and the first average value exceeds the first predetermined range, and the second area is An area in the image that is covered by the detection window; if the second preset range is exceeded, the second determining unit 504 is triggered,
  • the second determining unit 504 is configured to determine that the difference between the color value of each unit length in the second area and the first average value exceeds the number of the second preset range, if the number exceeds the The ratio between the number of unit lengths of the two preset ranges and the total number of unit lengths of the second area does not exceed a preset ratio, and the first area is determined to be a part of the silhouette of the portrait in the image.
  • the portrait wheel on the side of the moving direction of the detection window at the moving height of the image movement can be determined The position of the profile.
  • the detected train is used to traverse the image, and the above-described unit is repeatedly triggered during the traversal, thereby determining a complete portrait outline of the portrait in the image.
  • the moving unit 501 is further configured to: after triggering the second determining unit, translate the detection window to a width perpendicular to the horizontal moving direction, from a side of the image The other side moves the detection window horizontally, continuing to determine the position of the silhouette of the portrait in the image until the image is traversed through the detection window.
  • the detection window After determining the width of the detection window, when the detection window determines the position of a portrait contour, the detection window is returned to one side of the image, and the detection is moved up or down at the moving height. The width of the window continues to move the detection window to determine the position of the portrait of the portrait in the image at this moving height. In the process of traversing the image, it may be moved from one side to the other side, and then traversed in the reverse direction, for example, the detection window may be moved from the left edge to the right edge from the upper edge of the image, and the completion is completed.
  • the moving height of the detection window is moved downward, and after traversing the image from the top to the bottom, the positions of the determined portrait contours are stitched together to form a contour on the left side of the portrait in the image. Then, starting from the uppermost edge of the right side of the image, traversing the image from top to bottom in the direction from the right side to the left side of the image, after the traversal is completed, the determined silhouette of the portrait The positions are stitched together to form a contour to the right of the portrait in the image, and the previously determined left contour is combined to form a complete portrait of the portrait in the image.
  • the image is recorded on the image by the color detection area. a color value of the region; and during the moving of the detection window, continuously updating the first average value, in the process of moving the detection window, if a color value of the first region is between the first average value and If the difference exceeds the first preset range, stop moving the detection window, and determine whether the difference between the first average value and the second average value exceeds a second preset range, if the second preset range is exceeded, And the ratio between the number of unit lengths exceeding the second preset range and the total number of unit lengths of the second area does not exceed a preset ratio, and determining that the first area is a silhouette of a portrait in the image portion.
  • the detection is performed by moving the monitoring window.
  • the color value can be collected by moving horizontally from one side of the image to the other side, so that the images can be more accurately and effectively analyzed.
  • the change in the color value of the line results in a more accurate determination of the silhouette of the portrait.
  • the method further includes:
  • the determining unit is configured to determine that the first area is a pixel impurity in the image, and the color value of the first area is no longer used to determine a position of a silhouette of the portrait in the image.
  • it also includes:
  • the determination result of the second determining unit is that the ratio between the number of unit lengths exceeding the second preset range and the total number of unit lengths of the second area exceeds a preset ratio, triggering the movement
  • the unit continues to move the detection window and triggers the determination unit.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

本发明实施例公开了一种图像中人像轮廓的确定方法和装置，从图像的一侧向另一侧水平移动预设的检测窗口，记录在图像上被颜色检测区经过的区域的颜色值；在移动所述检测窗口的过程中，若第一区域的颜色值与所述第一平均值之间的差值超出第一预设范围，停止移动所述检测窗口，并判断第一平均值和第二平均值之间的差值是否超出第二预设范围，若超出所述第二预设范围，判断所述第二区域中各个单位长度的颜色值与所述第一平均值的差值超出所述第二预设范围的个数，若未超过预设比例，确定所述第一区域为所述图像中人像轮廓的一部分，通过从所述图像的一侧向另一侧水平移动来采集颜色值的方式，可以更准确有效的分析所述图像处于同一行的颜色值的变化。

Description

一种图像中人像轮廓的确定方法和装置
本申请要求2015年12月03日递交的申请号为201510883297.5、发明名称为“一种图像中人像轮廓的确定方法和装置”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明涉及图像处理领域，特别是涉及一种图像中人像轮廓的确定方法和装置。
背景技术
传统的形态学的轮廓判定方法常用来判定图像中包括的物体的轮廓。这里的物体可以主要包括建筑、人像等。确定出的物体轮廓可以用于后期的图像加工、处理等。
然而，目前这些针对物体轮廓的轮廓判定方法主要追求的是对轮廓判定的速度，即希望能够尽快的识别出物体的轮廓。而对物体轮廓识别准确性的关注度并不高。
针对主要包括人像的图像，例如证件照等，在后续处理过程中，对人像轮廓的识别精度要求较高。使用传统的物体轮廓的轮廓判定方法处理得到的人像轮廓，其人像轮廓的精度难以达到后续处理中对人像轮廓精度的要求。
发明内容
为了解决上述技术问题，本发明提供了一种图像中人像轮廓的确定方法和装置，通过移动检测窗口得到更为精确的人像轮廓的确定结果。
本发明实施例公开了如下技术方案：
一种图像中人像轮廓的确定方法，所述方法包括：
从图像的一侧向另一侧水平移动预设的检测窗口，所述检测窗口的长度为多个单位长度，其长度单位的个数根据所述图像预先确定，所述检测窗口包括一个单位长度的颜色检测区；
在移动所述检测窗口的过程中，记录在所述图像上被所述颜色检测区经过的区域的颜色值；并在移动所述检测窗口的过程中，持续更新第一平均值，所述第一平均值为所记录颜色值在单位长度下的颜色平均值；
在移动所述检测窗口的过程中，若第一区域的颜色值与所述第一平均值之间的差值超出第一预设范围，停止移动所述检测窗口，并判断第一平均值和第二平均值之间的差 值是否超出第二预设范围，其中，所述第一区域为所述图像中处于被所述颜色检测区覆盖状态的区域，第二平均值为当第一区域的颜色值与所述第一平均值之间的差值超出第一预设范围时，第二区域在单位长度下的颜色平均值，所述第二区域为所述图像中处于被所述检测窗口覆盖状态的区域；
若超出所述第二预设范围，判断所述第二区域中各个单位长度的颜色值与所述第一平均值的差值超出所述第二预设范围的个数，若超出所述第二预设范围的单位长度的个数与所述第二区域总单位长度个数之间的比例未超过预设比例，确定所述第一区域为所述图像中人像轮廓的一部分。
可选的，在所述确定所述第一区域为所述图像中人像轮廓的一部分之后，还包括：
将所述检测窗口沿着垂直于水平移动方向平移所述检测窗口的宽度，从所述图像的一侧向另一侧水平移动所述检测窗口，继续确定所述图像中人像轮廓的位置，直至通过所述检测窗口遍历完所述图像。
可选的，所述颜色检测区位于所述检测窗口中沿着水平移动方向的最后的位置。
可选的，所述判断第一平均值和第二平均值之间的差值是否超出第二预设范围，还包括：
若未超出所述第二预设范围，继续移动所述检测窗口，并判断所述第一区域为所述图像中的像素杂质，不再将所述第一区域的颜色值用于确定所述图像中人像轮廓的位置。
可选的，所述判断所述第二区域中各个单位长度的颜色值与所述第一平均值的差值超出所述第二预设范围的个数，还包括：
若超出所述第二预设范围的单位长度的个数与所述第二区域总单位长度个数之间的比例超过了预设比例，继续移动所述检测窗口，并判断所述第一区域为所述图像中的像素杂质，不再将所述第一区域的颜色值用于确定所述图像中人像轮廓的位置。
可选的，在所述从图像的一侧向另一侧水平移动预设的检测窗口之前，还包括：
将所述图像的颜色模型转换为色调饱和度亮度HSV颜色模型。
可选的，在所述从图像的一侧向另一侧水平移动预设的检测窗口之前，还包括：
通过人脸匹配，判断出所述图像中人像的人脸所处的矩形区域；
所述检测窗口的长度为多个单位长度，其长度单位的个数根据所述图像预先确定，具体包括：
根据所述矩形区域的形状参数确定得到所述检测窗口长度的单位长度的个数。
一种图像中人像轮廓的确定装置，所述装置包括：
移动单元，用于从图像的一侧向另一侧水平移动预设的检测窗口，所述检测窗口的长度为多个单位长度，其长度单位的个数根据所述图像预先确定，所述检测窗口包括一个单位长度的颜色检测区；
记录单元，用于在移动所述检测窗口的过程中，记录在所述图像上被所述颜色检测区经过的区域的颜色值；并在移动所述检测窗口的过程中，持续更新第一平均值，所述第一平均值为所记录颜色值在单位长度下的颜色平均值；
第一判断单元，用于在移动所述检测窗口的过程中，若第一区域的颜色值与所述第一平均值之间的差值超出第一预设范围，停止移动所述检测窗口，并判断第一平均值和第二平均值之间的差值是否超出第二预设范围，其中，所述第一区域为所述图像中处于被所述颜色检测区覆盖状态的区域，第二平均值为当第一区域的颜色值与所述第一平均值之间的差值超出第一预设范围时，第二区域在单位长度下的颜色平均值，所述第二区域为所述图像中处于被所述检测窗口覆盖状态的区域；若超出所述第二预设范围，触发第二判断单元，
所述第二判断单元，用于判断所述第二区域中各个单位长度的颜色值与所述第一平均值的差值超出所述第二预设范围的个数，若超出所述第二预设范围的单位长度的个数与所述第二区域总单位长度个数之间的比例未超过预设比例，确定所述第一区域为所述图像中人像轮廓的一部分。
可选的，
所述移动单元还用于在触发所述第二判断单元之后，将所述检测窗口沿着垂直于水平移动方向平移所述检测窗口的宽度，从所述图像的一侧向另一侧水平移动所述检测窗口，继续确定所述图像中人像轮廓的位置，直至通过所述检测窗口遍历完所述图像。
可选的，所述颜色检测区位于所述检测窗口中沿着水平移动方向的最后的位置。
可选的，还包括：
若所述第一判断单元的判断结果为未超出所述第二预设范围，触发所述移动单元继续移动所述检测窗口，并触发确定单元，
所述确定单元，用于判断所述第一区域为所述图像中的像素杂质，不再将所述第一区域的颜色值用于确定所述图像中人像轮廓的位置。
可选的，还包括：
若所述第二判断单元的判断结果为超出所述第二预设范围的单位长度的个数与所述第二区域总单位长度个数之间的比例超过了预设比例，触发所述移动单元继续移动所述 检测窗口，并触发所述确定单元。
可选的，还包括：
转换单元，用于在触发所述移动单元之前，将所述图像的颜色模型转换为色调饱和度亮度HSV颜色模型。
可选的，还包括：
匹配单元，用于在触发所述移动单元之前，通过人脸匹配，判断出所述图像中人像的人脸所处的矩形区域；
所述检测窗口的长度为多个单位长度，其长度单位的个数根据所述图像预先确定，具体包括：
根据所述矩形区域的形状参数确定得到所述检测窗口长度的单位长度的个数。
由上述技术方案可以看出，通过从图像的一侧向另一侧水平移动预设的检测窗口，在移动所述检测窗口的过程中，记录在所述图像上被所述颜色检测区经过的区域的颜色值；并在移动所述检测窗口的过程中，持续更新第一平均值，在移动所述检测窗口的过程中，若第一区域的颜色值与所述第一平均值之间的差值超出第一预设范围，停止移动所述检测窗口，并判断第一平均值和第二平均值之间的差值是否超出第二预设范围，若超出所述第二预设范围，且超出所述第二预设范围的单位长度的个数与所述第二区域总单位长度个数之间的比例未超过预设比例，确定所述第一区域为所述图像中人像轮廓的一部分。通过移动监测窗口进行检测，虽然需要采集处理的数据相对较多，但是通过从所述图像的一侧向另一侧水平移动来采集颜色值的方式，可以更准确有效的分析所述图像处于同一行的颜色值的变化，由此得到更为精确的人像轮廓的确定结果。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。
图1为本发明实施例提供的一种图像中人像轮廓的确定方法的方法流程图；
图2为本发明实施例提供的一种人脸检测区在图像中的示意图；
图3为本发明实施例提供的一种在图像中确定人像轮廓的示意图；
图4为本发明实施例提供的一种人像轮廓确定示意图；
图5为本发明实施例提供的一种图像中人像轮廓的确定装置的装置结构图。
具体实施方式
为使本发明实施例的目的、技术方案和优点更加清楚，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚地描述，显然，所描述的实施例是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
传统的形态学的轮廓判定方法常用来判定图像中包括的物体的轮廓。确定出的物体轮廓可以用于后期的图像加工、处理等。然而，目前这些针对物体轮廓的轮廓判定方法主要追求处理速度，希望能够尽可能快的识别出图像中的物体轮廓，然而追求速度下，导致物体轮廓识别的准确性不高。
针对一些主要包括人像的图像，例如证件照等，在后续处理过程中，例如更换证件照背景颜色、PS等，对人像轮廓的识别精度要求较高。目前传统的物体轮廓判定方法的轮廓判定精度不高，会出现对人像轮廓的边缘处理不清晰，或者造成人像的部分头发或有效部分被裁剪的情况发生。可见，传统轮廓判定方法的精度难以达到后期处理的需求。
为此，本发明实施例提供的一种图像中人像轮廓的确定方法和装置，通过从图像的一侧向另一侧水平移动预设的检测窗口，在移动所述检测窗口的过程中，记录在所述图像上被所述颜色检测区经过的区域的颜色值；并在移动所述检测窗口的过程中，持续更新第一平均值，在移动所述检测窗口的过程中，若第一区域的颜色值与所述第一平均值之间的差值超出第一预设范围，停止移动所述检测窗口，并判断第一平均值和第二平均值之间的差值是否超出第二预设范围，若超出所述第二预设范围，且超出所述第二预设范围的单位长度的个数与所述第二区域总单位长度个数之间的比例未超过预设比例，确定所述第一区域为所述图像中人像轮廓的一部分。通过移动监测窗口进行检测，虽然需要采集处理的数据相对较多，导致数据处理的时间较长，但是通过从所述图像的一侧向另一侧水平移动来采集颜色值的方式，可以更准确有效的分析所述图像处于同一行的颜色值的变化，由此得到更为精确的人像轮廓的确定结果。
实施例一
在描述本发明实施例提供的一种图像中人像轮廓的确定方法之前，先总体描述一下应用环境和实现机制。
本发明的执行主体可以是一个计算机程序，这个计算机程序可以安装部署在移动终 端、计算机、服务器等中。移动终端、计算机、服务器等通过运行部署在自身的这个计算机程序，能够实现本发明实施例所提供的对图像的图像处理流程，以确定出图像中人像的轮廓。
下文中出现的检测窗口，可以理解为本程序在图像上的颜色值的检测区间。例如，在启动本程序处理一个图像时，本程序在确定图像边缘后，确定出颜色值的检测区间(也就是检测窗口)，并检测落入这个颜色值的检测区间中图像部分的颜色值。获取到所需颜色值后，开始按照下文所述方式，在这个图像上移动这个颜色值的检测区间，并继续检测落入这个颜色值的检测区间中图像部分的颜色值。
下文中出现的颜色检测区可以理解为检测窗口中的一个特定的、固定的位置，或者说是这个颜色值的检测区间中的一个特定区域。在这个颜色值的检测区间移动过程中，颜色检测区在这个颜色值的检测区间中的位置相对固定。
本程序可以通过在内存或缓存等存储介质中，或者，也可以在专门划分出的一个存储区域中保存检测到的颜色值、计算出的平均值等参数，以便在图像处理需要这些参数时进行调用。
图1为本发明实施例提供的一种图像中人像轮廓的确定方法的方法流程图，所述方法包括：
S101：从图像的一侧向另一侧水平移动预设的检测窗口，所述检测窗口的长度为多个单位长度，其长度单位的个数根据所述图像预先确定，所述检测窗口包括一个单位长度的颜色检测区。
举例说明，本发明实施例中所述的图像，一般可以理解为具有浅色背景的人像图像。这里所述的浅色可以为白色、浅灰色等常见的背景颜色。人像图像可以为常见的半身证件照等。在从所述图像的一侧向另一侧水平移动预设的检测窗口之前，为了能够达到更好的图像处理效果，可以调整图像的颜色模型，可选的，将所述图像的颜色模型转换为色调饱和度亮度(英文：Hue,Saturation,Value，缩写：HSV)颜色模型。这种颜色模型相对于图像所采用的常见颜色模型，例如红绿蓝(英文：Red，Green，Blue，缩写：RGB)颜色模型来说，所采集的颜色值更为准确，所包含的信息更利于后续的图像处理。
由于检测窗口主要用于采集、处理所述图像中的像素数据(颜色值)，故所述检测窗口的形状和像素形状相关，具体可以为一个矩形的形状。对于所述检测窗口的长度为多个单位长度，其长度单位的个数根据所述图像预先确定，需要说明的是，根据具体的图像情况，在所述从图像的一侧向另一侧水平移动预设的检测窗口之前，本发明实施例 提供了至少两种设置所述检测窗口长度的方式。但是，需要注意的是，不论使用哪种方式设置所述检测窗口的长度，所述检测窗口的长度需要满足至少具有两个单位长度。本发明实施例中所述的单位长度和在图像处理中常用的、以像素为单位的设置具有对应的比例关系。具体设置一个单位长度等于几个像素，可以根据不同的应用场景或者人像轮廓确定精度来定。一般情况下，为了方便计算，一个单位长度可以等于整数个像素，例如可以等于一个像素，也可以等于两个像素。
第一种方式，对于常见的半身证件照来说，不论是一寸还是两寸的证件照，一般照片中人像的位置和大小能够符合一定尺寸比例，故针对这种类型的照片，所述检测窗口的长度可以根据所述照片(也就是所述图像)的尺寸进行设定。
第二种方式，可以应用于不同类型的具有人像的图像中，即通过所述图像中人脸所处的矩形区域的形状参数(例如宽度、长度)来设置所述检测窗口的长度。也就是可选的，通过人脸匹配，判断出所述图像中人像的人脸所处的矩形区域。S101中所述检测窗口的长度为多个单位长度，其长度单位的个数根据所述图像预先确定，具体包括：根据所述矩形区域的形状参数确定得到所述检测窗口长度的单位长度的个数。可以通过图像处理框架，做人脸匹配，本发明对此不限定。一般会确定出一个矩形区域又或者叫人脸检测区。其长宽都可以通过人脸匹配获得，长宽的长度单位一般是像素。具体可参见图2，图2为本发明实施例提供的一种人脸检测区在图像中的示意图。在图2中，W为人脸检测区的宽度，H为人脸检测区的长度，10为所述检测窗口。在图2中所示的是从所述图像的左侧向右侧移动所述检测窗口的情况。本发明实施例还提供了一种具体的根据所述矩形区域的形状参数确定得到所述检测窗口长度的单位长度的个数的计算方法，在一个单位长度等于一个像素的前提下，所述检测窗口的长度可以为人像检测区宽度的一半。具体计算方法为：L＝int(W/2)，其中，L为所述检测窗口的长度，int为取整标记，以免在计算人像检测区宽度的一半时出现除不尽的情况。
本发明并不限定颜色检测区在所述检测窗口中的位置，为了保证人像轮廓确定的准确性，仅限定所述颜色检测区的长度为一个单位长度。在本发明实施例提供的移动方式和计算方式下，一般来说，位于移动方向的尾部可以达到最佳的人像轮廓确定准确性。也就是说，可选的，所述颜色检测区可以位于所述检测窗口中沿着水平移动方向的最后的位置。如图2所示的11可以为从所述图像的左侧向右侧水平移动时，所述颜色检测区在所述检测窗口中的位置，即最左侧。若所述检测窗口从所述图像的右侧向左侧水平移动，所述颜色检测区处于所述检测窗口的最右侧。
从图像的一侧向另一侧水平移动的方式，由于是水平移动，故一般限定为从图像的左侧移动到右侧，或者从右侧移动到左侧，但是本发明实施例中并没有限定所述图像的摆放，所述图像可以如图2所示正摆放，即以照相时的正常方位摆放(一般来说人像的头像朝上)。所述图像也可以侧摆放，例如如图2所示的摆放角度顺时针或逆时针旋转90度的摆放方式，在这种侧摆放的情况下，从图像的一侧向另一侧水平移动的方式，则可以理解为从所述图像的下方水平移动到上方，或者从所述图像的上方水平移动到下方。
S102：在移动所述检测窗口的过程中，记录在所述图像上被所述颜色检测区经过的区域的颜色值；并在移动所述检测窗口的过程中，持续更新第一平均值，所述第一平均值为所记录颜色值在单位长度下的颜色平均值。
通过附图说明，图3为本发明实施例提供的一种在图像中确定人像轮廓的示意图，如图3所示，在所述图像上被所述颜色检测区经过的区域可以为所述颜色检测区11在移动过程中依次覆盖过的区域之和，图3中，每个虚线方格所划分的区域大小和所述颜色检测区11的大小相同，且长度均为一个单位长度。需要注意的是，在所述图像上被所述颜色检测区经过的区域包括图3所示虚线区域以及所述颜色检测区当前正覆盖的区域。需要注意的是，所记录的颜色值可以为所述颜色检测区经过的区域中各个像素的颜色值，在计算所述第一平均值时，不是以像素为单位计算，而是以单位长度为单位计算。假设在所述图像上被所述颜色检测区经过的区域共有100个像素，一个单位长度为两个像素，那么在步骤S102计算第一平均值时，是将100个像素的颜色值求和后，除以50(即100个像素为50个单位长度)。
S103：在移动所述检测窗口的过程中，若第一区域的颜色值与所述第一平均值之间的差值超出第一预设范围，停止移动所述检测窗口，并判断第一平均值和第二平均值之间的差值是否超出第二预设范围，其中，所述第一区域为所述图像中处于被所述颜色检测区覆盖状态的区域，第二平均值为当第一区域的颜色值与所述第一平均值之间的差值超出第一预设范围时，第二区域在单位长度下的颜色平均值，所述第二区域为所述图像中处于被所述检测窗口覆盖状态的区域。
举例说明，所述第一区域为所述图像中被所述颜色检测区覆盖的区域，所述第一区域会随着所述检测窗口在所述图像上的移动而移动。同理，所述第二区域也会随着所述检测窗口在所述图像上的移动而移动。具体位置可以参见图3所示。
由于在第一区域的颜色值与所述第一平均值之间的差值超出第一预设范围时停止移动所述检测窗口，那么所述第二平均值为所述检测窗口停止时，正覆盖的所述图像区域 中颜色值的平均值。在计算所述第二平均值时，仍然需要注意的是，不是以像素为单位计算，而是以单位长度为单位计算。
需要注意的是，所述第一预设范围和所述第二预设范围可以相同，也可以不同，例如可以均为±20％。通过设置第一预设范围和第二预设范围，可以用于判断所述图像在所述检测窗口移动路径上的颜色值的突变，且这个突变的点处于所述第一区域中。当然，这种突变有可能是所述图像上作为人像背景中的杂质，也有可能是所述图像中人像轮廓的位置，具体是哪一种可能，可以通过S104中的判断进行有效的区分和识别。
S104：若超出所述第二预设范围，判断所述第二区域中各个单位长度的颜色值与所述第一平均值的差值超出所述第二预设范围的个数，若超出所述第二预设范围的单位长度的个数与所述第二区域总单位长度个数之间的比例未超过预设比例，确定所述第一区域为所述图像中人像轮廓的一部分。
举例说明，通过判断第二区域中颜色值超出所述第一平均值达到第二预设范围的单位长度占第二区域总单位长度的比例，可以确定所述第二区域中颜色值变化的趋势是否平稳，从而判断目前所述检测窗口是否进入了所述图像中人像的区域内。所述预设比例可以根据具体应用场景或者判断人像轮廓的准确度要求进行设置，一般设置一个较小的值，例如25％。
在S104的判断中，若超出所述第二预设范围的单位长度的个数与所述第二区域总单位长度个数之间的比例未超过预设比例，那么证明所述检测窗口已经进入到所述图像中人像的区域内。这种情况可以参见图4，图4为本发明实施例提供的一种人像轮廓确定示意图。可见，当所述检测窗口进入所述人像区域内，且所述颜色检测区正位于所述人像轮廓上时，通过S104中的判断可以识别出这一情况。
通过S101至S104的方案，可以确定出所述检测窗口在所述图像移动的移动高度上移动方向一侧的人像轮廓的位置。之后，通过改变所述检测窗口的移动高度和移动方向，使用所述检测列车遍历所述图像，在遍历的过程中重复S101至S104的方案，从而确定出所述图像中人像的完整人像轮廓。
可选的，在所述确定所述第一区域为所述图像中人像轮廓的一部分之后，还包括：
将所述检测窗口沿着垂直于水平移动方向平移所述检测窗口的宽度，从所述图像的一侧向另一侧水平移动所述检测窗口，继续确定所述图像中人像轮廓的位置，直至通过所述检测窗口遍历完所述图像。
在这里说明所述检测窗口的宽度，一般来说，为了尽可能的提高检测人像轮廓的准 确性，希望将所述检测窗口设置的尽量的窄，也就是说，所述检测窗口越窄，使用所述检测窗口确定所述图像中人像的人像轮廓的准确性就越高。本发明实施例提供一种可选的所述检测窗口的宽度，也就是一个像素。当然也可以根据具体场景的检测精度要求，适应性的调整所述检测窗口的宽度。
在确定了所述检测窗口的宽度后，在所述检测窗口确定出一个人像轮廓的位置时，将所述检测窗口退回所述图像的一侧，在移动高度上上移或下移所述检测窗口的宽度，并继续移动所述检测窗口来确定这个移动高度上的所述图像中人像的人像轮廓的位置。在遍历所述图像的过程中，可以先从一侧向另一侧移动，之后在反向遍历，例如可以从所述图像的最上沿，从左侧到右侧开始移动所述检测窗口，完成一次检测后，下移所述检测窗口的移动高度，从左侧，从上至下遍历完所述图像后，将确定出的人像轮廓的位置拼接在一起可以形成所述图像中人像左边的轮廓，接下来，可以从所述图像的右侧的最上沿开始，从所述图像右侧到左侧的方向，从上至下开始遍历所述图像，遍历完成后，将确定出的人像轮廓的位置拼接在一起可以形成所述图像中人像右边的轮廓，和之前确定的左侧的轮廓组合在一起形成所述图像中人像的完整人像轮廓。
由上述实施例可以看出，通过从图像的一侧向另一侧水平移动预设的检测窗口，在移动所述检测窗口的过程中，记录在所述图像上被所述颜色检测区经过的区域的颜色值；并在移动所述检测窗口的过程中，持续更新第一平均值，在移动所述检测窗口的过程中，若第一区域的颜色值与所述第一平均值之间的差值超出第一预设范围，停止移动所述检测窗口，并判断第一平均值和第二平均值之间的差值是否超出第二预设范围，若超出所述第二预设范围，且超出所述第二预设范围的单位长度的个数与所述第二区域总单位长度个数之间的比例未超过预设比例，确定所述第一区域为所述图像中人像轮廓的一部分。通过移动监测窗口进行检测，虽然需要采集处理的数据相对较多，但是通过从所述图像的一侧向另一侧水平移动来采集颜色值的方式，可以更准确有效的分析所述图像处于同一行的颜色值的变化，由此得到更为精确的人像轮廓的确定结果。
实施例二
在本实施例中，将主要描述通过本发明实施例提供的方法，如何确定出所述图像上作为人像背景中的杂质或不合格的点。
在图1所对应实施例中的S103中，所述判断第一平均值和第二平均值之间的差值是否超出第二预设范围，还包括：
若未超出所述第二预设范围，继续移动所述检测窗口，并判断所述第一区域为所述图像中的像素杂质，不再将所述第一区域的颜色值用于确定所述图像中人像轮廓的位置。
举例说明，若所述第一平均值和第二平均值之间的差值未超出所述第二预设范围，可以确定所述第二区域中颜色值变化与在所述图像上被所述颜色检测区经过的区域的颜色值变化之间差别不大，那么所述检测窗口可能并没有完全进入所述图像中人像的区域内，那么，所述检测窗口中的颜色检测区所覆盖的位置并不能作为所述图像中人像的人像轮廓。故可以确定所述第一区域为所述图像上作为人像背景中的杂质，或者说是所述图像中不合格的点。这种杂质点或不合格的点会影响到之后人像轮廓确定的准确性，故为了保证准确性，不再将所述第一区域的颜色值用于确定所述图像中人像轮廓的位置。例如可以在S102中记录的颜色值中删除掉这个颜色值等。
可选的，所述判断所述第二区域中各个单位长度的颜色值与所述第一平均值的差值超出所述第二预设范围的个数，还包括：
若超出所述第二预设范围的单位长度的个数与所述第二区域总单位长度个数之间的比例超过了预设比例，继续移动所述检测窗口，并判断所述第一区域为所述图像中的像素杂质，不再将所述第一区域的颜色值用于确定所述图像中人像轮廓的位置。
举例说明，若超出所述第二预设范围的单位长度的个数与所述第二区域总单位长度个数之间的比例超过了预设比例，可以确定所述第二区域中颜色值变化的趋势并不平稳，那么所述检测窗口可能并没有完全进入所述图像中人像的区域内，那么，所述检测窗口中的颜色检测区所覆盖的位置并不能作为所述图像中人像的人像轮廓。但是，由于第一平均值和第二平均值之间的差值已经超出了第二预设范围，第一区域的颜色值与所述第一平均值之间的差值也已经超出第一预设范围，所以可以确定的是，所述第一区域中的像素点的颜色值出现了突然的改变。故可以确定所述第一区域为所述图像上作为人像背景中的杂质，或者说是所述图像中不合格的点。这种杂质点或不合格的点会影响到之后人像轮廓确定的准确性，故为了保证准确性，不再将所述第一区域的颜色值用于确定所述图像中人像轮廓的位置。例如可以在S102中记录的颜色值中删除掉这个颜色值等。
由上述实施例可以看出，通过识别所述图像中的像素杂质或不合格的点，并不再将该杂质或不合格的点的颜色值用于确定所述图像中人像轮廓的位置的过程中，由此可以有效的提高确定人像轮廓位置的准确性。
实施例三
图5为本发明实施例提供的一种图像中人像轮廓的确定装置的装置结构图，所述装置包括：
移动单元501，用于从图像的一侧向另一侧水平移动预设的检测窗口，所述检测窗口的长度为多个单位长度，其长度单位的个数根据所述图像预先确定，所述检测窗口包括一个单位长度的颜色检测区。
可选的，还包括：
转换单元，用于在触发所述移动单元之前，将所述图像的颜色模型转换为色调饱和度亮度HSV颜色模型。
可选的，还包括：
匹配单元，用于在触发所述移动单元之前，通过人脸匹配，判断出所述图像中人像的人脸所处的矩形区域；
所述检测窗口的长度为多个单位长度，其长度单位的个数根据所述图像预先确定，具体包括：
根据所述矩形区域的形状参数确定得到所述检测窗口长度的单位长度的个数。
可选的，所述颜色检测区位于所述检测窗口中沿着水平移动方向的最后的位置。
记录单元502，用于在移动所述检测窗口的过程中，记录在所述图像上被所述颜色检测区经过的区域的颜色值；并在移动所述检测窗口的过程中，持续更新第一平均值，所述第一平均值为所记录颜色值在单位长度下的颜色平均值。
第一判断单元503，用于在移动所述检测窗口的过程中，若第一区域的颜色值与所述第一平均值之间的差值超出第一预设范围，停止移动所述检测窗口，并判断第一平均值和第二平均值之间的差值是否超出第二预设范围，其中，所述第一区域为所述图像中处于被所述颜色检测区覆盖状态的区域，第二平均值为当第一区域的颜色值与所述第一平均值之间的差值超出第一预设范围时，第二区域在单位长度下的颜色平均值，所述第二区域为所述图像中处于被所述检测窗口覆盖状态的区域；若超出所述第二预设范围，触发第二判断单元504，
所述第二判断单元504，用于判断所述第二区域中各个单位长度的颜色值与所述第一平均值的差值超出所述第二预设范围的个数，若超出所述第二预设范围的单位长度的个数与所述第二区域总单位长度个数之间的比例未超过预设比例，确定所述第一区域为所述图像中人像轮廓的一部分。
由此可以确定出所述检测窗口在所述图像移动的移动高度上移动方向一侧的人像轮 廓的位置。之后，通过改变所述检测窗口的移动高度和移动方向，使用所述检测列车遍历所述图像，在遍历的过程中重复触发上述单元，从而确定出所述图像中人像的完整人像轮廓。
可选的，所述移动单元501还用于在触发所述第二判断单元之后，将所述检测窗口沿着垂直于水平移动方向平移所述检测窗口的宽度，从所述图像的一侧向另一侧水平移动所述检测窗口，继续确定所述图像中人像轮廓的位置，直至通过所述检测窗口遍历完所述图像。
在确定了所述检测窗口的宽度后，在所述检测窗口确定出一个人像轮廓的位置时，将所述检测窗口退回所述图像的一侧，在移动高度上上移或下移所述检测窗口的宽度，并继续移动所述检测窗口来确定这个移动高度上的所述图像中人像的人像轮廓的位置。在遍历所述图像的过程中，可以先从一侧向另一侧移动，之后在反向遍历，例如可以从所述图像的最上沿，从左侧到右侧开始移动所述检测窗口，完成一次检测后，下移所述检测窗口的移动高度，从左侧，从上至下遍历完所述图像后，将确定出的人像轮廓的位置拼接在一起可以形成所述图像中人像左边的轮廓，接下来，可以从所述图像的右侧的最上沿开始，从所述图像右侧到左侧的方向，从上至下开始遍历所述图像，遍历完成后，将确定出的人像轮廓的位置拼接在一起可以形成所述图像中人像右边的轮廓，和之前确定的左侧的轮廓组合在一起形成所述图像中人像的完整人像轮廓。
由上述实施例可以看出，通过从图像的一侧向另一侧水平移动预设的检测窗口，在移动所述检测窗口的过程中，记录在所述图像上被所述颜色检测区经过的区域的颜色值；并在移动所述检测窗口的过程中，持续更新第一平均值，在移动所述检测窗口的过程中，若第一区域的颜色值与所述第一平均值之间的差值超出第一预设范围，停止移动所述检测窗口，并判断第一平均值和第二平均值之间的差值是否超出第二预设范围，若超出所述第二预设范围，且超出所述第二预设范围的单位长度的个数与所述第二区域总单位长度个数之间的比例未超过预设比例，确定所述第一区域为所述图像中人像轮廓的一部分。通过移动监测窗口进行检测，虽然需要采集处理的数据相对较多，但是通过从所述图像的一侧向另一侧水平移动来采集颜色值的方式，可以更准确有效的分析所述图像处于同一行的颜色值的变化，由此得到更为精确的人像轮廓的确定结果。
接下来描述如何确定出所述图像上作为人像背景中的杂质或不合格的点。
在图5所对应实施例的基础上，可选的，还包括：
若所述第一判断单元的判断结果为未超出所述第二预设范围，触发所述移动单元继 续移动所述检测窗口，并触发确定单元，
所述确定单元，用于判断所述第一区域为所述图像中的像素杂质，不再将所述第一区域的颜色值用于确定所述图像中人像轮廓的位置。
可选的，还包括：
若所述第二判断单元的判断结果为超出所述第二预设范围的单位长度的个数与所述第二区域总单位长度个数之间的比例超过了预设比例，触发所述移动单元继续移动所述检测窗口，并触发所述确定单元。
可见，通过识别所述图像中的像素杂质或不合格的点，并不再将该杂质或不合格的点的颜色值用于确定所述图像中人像轮廓的位置的过程中，由此可以有效的提高确定人像轮廓位置的准确性。
本领域普通技术人员可以理解：实现上述方法实施例的全部或部分步骤可以通过程序指令相关的硬件来完成，前述程序可以存储于一计算机可读取存储介质中，该程序在执行时，执行包括上述方法实施例的步骤；而前述的存储介质可以是下述介质中的至少一种：只读存储器(英文：read-only memory，缩写：ROM)、RAM、磁碟或者光盘等各种可以存储程序代码的介质。
需要说明的是，本说明书中的各个实施例均采用递进的方式描述，各个实施例之间相同相似的部分互相参见即可，每个实施例重点说明的都是与其他实施例的不同之处。尤其，对于设备及系统实施例而言，由于其基本相似于方法实施例，所以描述得比较简单，相关之处参见方法实施例的部分说明即可。以上所描述的设备及系统实施例仅仅是示意性的，其中作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。本领域普通技术人员在不付出创造性劳动的情况下，即可以理解并实施。
以上所述，仅为本发明较佳的具体实施方式，但本发明的保护范围并不局限于此，任何熟悉本技术领域的技术人员在本发明揭露的技术范围内，可轻易想到的变化或替换，都应涵盖在本发明的保护范围之内。因此，本发明的保护范围应该以权利要求的保护范围为准。

Claims (14)

  1. 一种图像中人像轮廓的确定方法，其特征在于，所述方法包括：
    从图像的一侧向另一侧水平移动预设的检测窗口，所述检测窗口的长度为多个单位长度，其长度单位的个数根据所述图像预先确定，所述检测窗口包括一个单位长度的颜色检测区；
    在移动所述检测窗口的过程中，记录在所述图像上被所述颜色检测区经过的区域的颜色值；并在移动所述检测窗口的过程中，持续更新第一平均值，所述第一平均值为所记录颜色值在单位长度下的颜色平均值；
    在移动所述检测窗口的过程中，若第一区域的颜色值与所述第一平均值之间的差值超出第一预设范围，停止移动所述检测窗口，并判断第一平均值和第二平均值之间的差值是否超出第二预设范围，其中，所述第一区域为所述图像中处于被所述颜色检测区覆盖状态的区域，第二平均值为当第一区域的颜色值与所述第一平均值之间的差值超出第一预设范围时，第二区域在单位长度下的颜色平均值，所述第二区域为所述图像中处于被所述检测窗口覆盖状态的区域；
    若超出所述第二预设范围，判断所述第二区域中各个单位长度的颜色值与所述第一平均值的差值超出所述第二预设范围的个数，若超出所述第二预设范围的单位长度的个数与所述第二区域总单位长度个数之间的比例未超过预设比例，确定所述第一区域为所述图像中人像轮廓的一部分。
  2. 根据权利要求1所述的方法，其特征在于，在所述确定所述第一区域为所述图像中人像轮廓的一部分之后，还包括：
    将所述检测窗口沿着垂直于水平移动方向平移所述检测窗口的宽度，从所述图像的一侧向另一侧水平移动所述检测窗口，继续确定所述图像中人像轮廓的位置，直至通过所述检测窗口遍历完所述图像。
  3. 根据权利要求1所述的方法，其特征在于，所述颜色检测区位于所述检测窗口中沿着水平移动方向的最后的位置。
  4. 根据权利要求1至3任一项所述的方法，其特征在于，所述判断第一平均值和第二平均值之间的差值是否超出第二预设范围，还包括：
    若未超出所述第二预设范围，继续移动所述检测窗口，并判断所述第一区域为所述图像中的像素杂质，不再将所述第一区域的颜色值用于确定所述图像中人像轮廓的位置。
  5. 根据权利要求1至3任一项所述的方法，其特征在于，所述判断所述第二区域中各个单位长度的颜色值与所述第一平均值的差值超出所述第二预设范围的个数，还包括：
    若超出所述第二预设范围的单位长度的个数与所述第二区域总单位长度个数之间的比例超过了预设比例，继续移动所述检测窗口，并判断所述第一区域为所述图像中的像素杂质，不再将所述第一区域的颜色值用于确定所述图像中人像轮廓的位置。
  6. 根据权利要求1所述的方法，其特征在于，在所述从图像的一侧向另一侧水平移动预设的检测窗口之前，还包括：
    将所述图像的颜色模型转换为色调饱和度亮度HSV颜色模型。
  7. 根据权利要求1所述的方法，其特征在于，在所述从图像的一侧向另一侧水平移动预设的检测窗口之前，还包括：
    通过人脸匹配，判断出所述图像中人像的人脸所处的矩形区域；
    所述检测窗口的长度为多个单位长度，其长度单位的个数根据所述图像预先确定，具体包括：
    根据所述矩形区域的形状参数确定得到所述检测窗口长度的单位长度的个数。
  8. 一种图像中人像轮廓的确定装置，其特征在于，所述装置包括：
    移动单元，用于从图像的一侧向另一侧水平移动预设的检测窗口，所述检测窗口的长度为多个单位长度，其长度单位的个数根据所述图像预先确定，所述检测窗口包括一个单位长度的颜色检测区；
    记录单元，用于在移动所述检测窗口的过程中，记录在所述图像上被所述颜色检测区经过的区域的颜色值；并在移动所述检测窗口的过程中，持续更新第一平均值，所述第一平均值为所记录颜色值在单位长度下的颜色平均值；
    第一判断单元，用于在移动所述检测窗口的过程中，若第一区域的颜色值与所述第一平均值之间的差值超出第一预设范围，停止移动所述检测窗口，并判断第一平均值和第二平均值之间的差值是否超出第二预设范围，其中，所述第一区域为所述图像中处于被所述颜色检测区覆盖状态的区域，第二平均值为当第一区域的颜色值与所述第一平均值之间的差值超出第一预设范围时，第二区域在单位长度下的颜色平均值，所述第二区域为所述图像中处于被所述检测窗口覆盖状态的区域；若超出所述第二预设范围，触发第二判断单元，
    所述第二判断单元，用于判断所述第二区域中各个单位长度的颜色值与所述第一平 均值的差值超出所述第二预设范围的个数，若超出所述第二预设范围的单位长度的个数与所述第二区域总单位长度个数之间的比例未超过预设比例，确定所述第一区域为所述图像中人像轮廓的一部分。
  9. 根据权利要求8所述的装置，其特征在于，
    所述移动单元还用于在触发所述第二判断单元之后，将所述检测窗口沿着垂直于水平移动方向平移所述检测窗口的宽度，从所述图像的一侧向另一侧水平移动所述检测窗口，继续确定所述图像中人像轮廓的位置，直至通过所述检测窗口遍历完所述图像。
  10. 根据权利要求8所述的装置，其特征在于，所述颜色检测区位于所述检测窗口中沿着水平移动方向的最后的位置。
  11. 根据权利要求8至10任一项所述的装置，其特征在于，还包括：
    若所述第一判断单元的判断结果为未超出所述第二预设范围，触发所述移动单元继续移动所述检测窗口，并触发确定单元，
    所述确定单元，用于判断所述第一区域为所述图像中的像素杂质，不再将所述第一区域的颜色值用于确定所述图像中人像轮廓的位置。
  12. 根据权利要求8至10任一项所述的装置，其特征在于，还包括：
    若所述第二判断单元的判断结果为超出所述第二预设范围的单位长度的个数与所述第二区域总单位长度个数之间的比例超过了预设比例，触发所述移动单元继续移动所述检测窗口，并触发所述确定单元。
  13. 根据权利要求8所述的装置，其特征在于，还包括：
    转换单元，用于在触发所述移动单元之前，将所述图像的颜色模型转换为色调饱和度亮度HSV颜色模型。
  14. 根据权利要求8所述的装置，其特征在于，还包括：
    匹配单元，用于在触发所述移动单元之前，通过人脸匹配，判断出所述图像中人像的人脸所处的矩形区域；
    所述检测窗口的长度为多个单位长度，其长度单位的个数根据所述图像预先确定，具体包括：
    根据所述矩形区域的形状参数确定得到所述检测窗口长度的单位长度的个数。
PCT/CN2016/106850 2015-12-03 2016-11-23 一种图像中人像轮廓的确定方法和装置 WO2017092589A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201510883297.5 2015-12-03
CN201510883297.5A CN106846419B (zh) 2015-12-03 2015-12-03 一种图像中人像轮廓的确定方法和装置

Publications (1)

Publication Number Publication Date
WO2017092589A1 true WO2017092589A1 (zh) 2017-06-08

Family

ID=58796270

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/106850 WO2017092589A1 (zh) 2015-12-03 2016-11-23 一种图像中人像轮廓的确定方法和装置

Country Status (2)

Country Link
CN (1) CN106846419B (zh)
WO (1) WO2017092589A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112991470A (zh) * 2021-02-08 2021-06-18 上海通办信息服务有限公司 一种复杂背景下的证件寸照背景颜色检查方法及其系统

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107590838B (zh) * 2017-08-18 2021-08-17 陕西维视智造科技股份有限公司 一种金属表面颜色视觉检测系统

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1145674A (zh) * 1994-12-30 1997-03-19 菲利浦电子有限公司 数字乳房x射线照片中的自动分割及轮廓线和乳头的检测
CN101013444A (zh) * 2007-02-13 2007-08-08 华为技术有限公司 一种自适应生成足球视频摘要的方法和装置
US20100054609A1 (en) * 2008-08-26 2010-03-04 Oki Data Corporation Image processing apparatus
CN101923637A (zh) * 2010-07-21 2010-12-22 康佳集团股份有限公司 一种移动终端及其人脸检测方法和装置
CN101996310A (zh) * 2009-08-12 2011-03-30 Tcl数码科技（深圳）有限责任公司 基于嵌入式系统的人脸检测与跟踪方法
CN103179332A (zh) * 2011-12-22 2013-06-26 洛阳理工学院 一种基于fpga技术的视觉目标自适应检测控制器
CN103208005A (zh) * 2012-01-13 2013-07-17 富士通株式会社 对象识别方法和对象识别设备

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101763504B (zh) * 2009-12-30 2012-06-06 邱怀骏 复杂场景下的人体头部识别方法
CN104636749A (zh) * 2013-11-14 2015-05-20 中国移动通信集团公司 目标对象检测方法及装置

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN1145674A (zh) * 1994-12-30 1997-03-19 菲利浦电子有限公司 数字乳房x射线照片中的自动分割及轮廓线和乳头的检测
CN101013444A (zh) * 2007-02-13 2007-08-08 华为技术有限公司 一种自适应生成足球视频摘要的方法和装置
US20100054609A1 (en) * 2008-08-26 2010-03-04 Oki Data Corporation Image processing apparatus
CN101996310A (zh) * 2009-08-12 2011-03-30 Tcl数码科技（深圳）有限责任公司 基于嵌入式系统的人脸检测与跟踪方法
CN101923637A (zh) * 2010-07-21 2010-12-22 康佳集团股份有限公司 一种移动终端及其人脸检测方法和装置
CN103179332A (zh) * 2011-12-22 2013-06-26 洛阳理工学院 一种基于fpga技术的视觉目标自适应检测控制器
CN103208005A (zh) * 2012-01-13 2013-07-17 富士通株式会社 对象识别方法和对象识别设备

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112991470A (zh) * 2021-02-08 2021-06-18 上海通办信息服务有限公司 一种复杂背景下的证件寸照背景颜色检查方法及其系统
CN112991470B (zh) * 2021-02-08 2023-12-26 上海通办信息服务有限公司 一种复杂背景下的证件寸照背景颜色检查方法及其系统

Also Published As

Publication number Publication date
CN106846419B (zh) 2020-06-05
CN106846419A (zh) 2017-06-13

Similar Documents

Publication Publication Date Title
JP6336117B2 (ja) 建物高さの計算方法、装置及び記憶媒体
CN105005766B (zh) 一种车身颜色识别方法
TW202011733A (zh) 對影像進行目標取樣的方法及裝置
KR101683704B1 (ko) 얼굴 보정 방법, 시스템 및 컴퓨터 저장 매체
US20190311470A1 (en) Apparel production monitoring system using image recognition
CN105744268A (zh) 摄像头遮挡检测方法及装置
US20200250840A1 (en) Shadow detection method and system for surveillance video image, and shadow removing method
JP2006350558A5 (zh)
CN107347151A (zh) 双目摄像头遮挡检测方法及装置
JP2006350559A5 (zh)
JP6582516B2 (ja) 画像におけるカラーチャートの検出装置及び検出方法
CN106067177A (zh) Hdr场景侦测方法和装置
JP7024736B2 (ja) 画像処理装置、および画像処理方法、並びにプログラム
KR100857463B1 (ko) 포토프린팅을 위한 얼굴영역 검출장치 및 보정 방법
EP2813973A1 (en) Method and system for processing video image
JP6197340B2 (ja) 画像処理装置、画像処理方法、およびプログラム
US8482630B2 (en) Apparatus and method for adjusting automatic white balance by detecting effective area
WO2017092589A1 (zh) 一种图像中人像轮廓的确定方法和装置
CN109035331B (zh) 一种信号灯组的位置矫正方法及设备
TWI530913B (zh) 移動物體偵測系統及方法
CN112312108B (zh) 一种白平衡的异常确定方法、装置、存储介质及电子设备
CN105809677B (zh) 一种基于双边滤波器的图像边缘检测方法及系统
WO2017000827A1 (zh) 一种自动筛选图片的方法及装置
JP2012226676A (ja) 棒材の計数方法
CN107273804A (zh) 基于支持向量机和深度特征的行人识别方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 16869899

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 16869899

Country of ref document: EP

Kind code of ref document: A1