Classifications

• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T7/00—Image analysis
  • G06T7/10—Segmentation; Edge detection
  • G06T7/12—Edge-based segmentation
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
  • G06V10/00—Arrangements for image or video recognition or understanding
  • G06V10/40—Extraction of image or video features
  • G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; 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
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2207/00—Indexing scheme for image analysis or image enhancement
  • G06T2207/10—Image acquisition modality
  • G06T2207/10016—Video; Image sequence
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; COUNTING
  • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
  • G06T2207/00—Indexing scheme for image analysis or image enhancement
  • G06T2207/20—Special algorithmic details
  • G06T2207/20092—Interactive image processing based on input by user
  • G06T2207/20096—Interactive definition of curve of interest
• • G—PHYSICS
  • G06—COMPUTING OR CALCULATING; 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 a digital image processing method for automatically outlining a contour of a figure in a digital image. Specifically, the present invention relates to employing a contour outlining tool with region detectors and a cost map to outline a contour of a figure in a digital image BACKGROUND OF THE INVENTION Detection of people in images or videos has been of interest in recent years in computer vision, image understanding, and image processing communities for various applications. The problem of detecting people is a very challenging one due to the fact that there are practically unlimited variations in configurations and appearances of human figures, particularly in terms of color, posture, and texture. In the past, researchers have explored people detection mainly using motion information or explicit models (see “Model-Based Vision: A Program to See a Walking Person," by David Hogg, Image and Vision Computing, Vol. 1, No. 1, February 1983, pp. 5-20).
• the Mortensen and Barrett scheme is basically a semi-automatic object contour extraction method that allows a user to enter his knowledge of the image and leave the algorithm to quickly and accurately extract object boundaries of interest.
• the Intelligent Scissors can be regarded as a 'non-parametric' boundary finding algorithm.
• Snakes were introduced as an energy minimizing splines guided by external and internal forces. Snakes are interactively initialized with an approximate boundary contour; subsequently this single contour is iteratively adjusted in an attempt to minimize an energy functional.
• snakes are curves that minimize an energy functional with a term for the internal energy of the curve and a term giving external forces to the curve.
• the internal energy consists of curve bending energy and curve stretching energy.
• the external forces are linked to the gradient of the image causing the snake to be drawn to the image's edges.
• snakes are numerical solutions of the Euler equations for a functional minimization problem. In the work done by Flikner et al.
• the original Intelligent Scissors tool is interactively initialized with just a single seed point and it then generates, at interactive speeds, all possible optimal paths from the seed to every other point in the image. Thus, the user is allowed to interactively select the desired optimal boundary segment.
• Intelligent Scissors typically requires less time and effort to segment an object than it takes to manually input an initial approximation to an object boundary.
• the Intelligent Scissors outperforms the snakes in outlining object boundaries having arbitrary shapes.
• a digital image processing method for automatically outlining a contour of a figure in a digital image including: testing parameters of a region within the digital image according to a plurality of cascaded tests; determining whether the region contains characteristic features of the figure within the digital image; computing location parameters of the characteristic features in the region for the figure within the digital image; determining boundary parameters for the figure corresponding to the location parameters of the characteristic features in the region; computing an information map of the digital image; computing a set of indicative pixels for the contour of the figure; and automatically outlining the contour of the figure using the set of indicative pixels, the information map, and a contour outlining tool.
• ADVANTAGES ADVANTAGES
• the present invention has the advantage of automatically finding the boundary of a figure in an image without user input.
• Fig. 1 is a schematic diagram of an image processing system useful in practicing the present invention
• Fig. 2 is a flowchart illustrating the face detection method of the present invention
• Fig. 3 is a flowchart illustrating the eye detection method of the present invention
• Fig. 4 is a flowchart illustrating the human figure contour outlining method according to the present invention.
• Fig. 5 is an illustration of several parameters for the people detection method according to the present invention.
• Fig. 6 is an illustration of a cost map along with search windows for searching seed and target pixel pairs according the present invention.
• Fig. 7 is a flowchart illustrating the figure contour outlining method according to the present invention. DETAILED DESCRIPTION OF THE INVENTION
• FIG. 1 shows an image processing system useful in practicing the present invention including a color digital image source 100, such as a film scanner, digital camera, or digital image storage device such as a compact disk drive with a Picture CD, or other image storage devices for digital images, such as servers, or wireless devices.
• the digital image from the digital image source 100 is provided to an image processor 102, such as a programmable personal computer, or digital image processing work station such as a Sun SparcTM workstation.
• the image processor 102 may be connected to a CRT image display 104, an operator interface such as a keyboard 106 and a mouse 108.
• Image processor 102 is also connected to computer readable storage medium 107.
• the image processor 102 transmits processed digital images to an output device 109.
• Output device 109 can comprise a hard copy printer, a long-term image storage device, a connection to another processor, or an image telecommunication device connected, for example, to the Internet, or a wireless device.
• the present invention comprises a computer program product for detecting human faces in a digital image in accordance with the method described.
• the computer program of the present invention can be utilized by any well-known computer system, such as the personal computer of the type shown in FIG. 1.
• many other types of computer systems can be used to execute the computer program of the present invention.
• the method of the present invention can be executed in the computer contained in a digital camera or a device combined or inclusive with a digital camera. Consequently, the computer system will not be discussed in further detail herein.
• the computer program product of the present invention may make use of image manipulation algorithms and processes that are well known. Accordingly, the present description will be directed in particular to those algorithms and processes forming part of, or cooperating more directly with, the method of the present invention.
• the computer program product embodiment of the present invention may embody algorithms and processes not specifically shown or described herein that are useful for implementation. Such algorithms and processes are conventional and within the ordinary skill in such arts. Other aspects of such algorithms and systems, and hardware and/or software for producing and otherwise processing the images involved or cooperating with the computer program product of the present invention, are not specifically shown or described herein and may be selected from such algorithms, systems, hardware, components, and elements known in the art.
• the computer program for performing the method of the present invention may be stored in a computer readable storage medium.
• This medium may comprise, for example: magnetic storage media such as a magnetic disk (such as a hard drive or a floppy disk) or magnetic tape; optical storage media such as an optical disc, optical tape, or machine readable bar code; solid state electronic storage devices such as random access memory (RAM), or read only memory (ROM); or any other physical device or medium employed to store a computer program.
• the computer program for performing the method of the present invention may also be stored on computer readable storage medium that is connected to the image processor by way of the Internet or other communication medium. Those skilled in the art will readily recognize that the equivalent of such a computer program product may also be constructed in hardware.
• FIG. 7 is a flowchart 700 illustrating the general figure contour outlining method according to the present invention.
• a first task is to find the figure or figures in the image.
• a second task is to outline contours of the figures.
• the first task in this method is accomplished by using a region parameter test 702, a characteristic feature determination 704 and a figure detection process 706.
• the region parameter test 702 finds a part or parts of a figure through methods such as template matching or correlation testing. Templates of the part or parts of the figure can be trained through algorithms such as supervised statistical learning.
• the characteristic feature determination 704 extracts distinct point or points from the tested region. The point or points have statistically meaningful relation to the figure to be outlined.
• the edges of the figure(s) can also be obtained through background and foreground (that is, the figures) separation using techniques such as color segmentation.
• the exemplary gradient operation and color segmentation could be intermediate results of the operation of information map generator 708.
• the information map generator 708 could be computed, for example, using color entities.
• the output of the information map generator 708 is a cost map or an energy map on which the operation of figure outlining 712 relies.
• the cost map could be computed using the gradient of the intensity values of the pixels of the image. With the cost map, the figure outlining operation 712 finds the minimum cost path from the seed to the target for a pair of seed and target points found in seed/target pixel pairs generation 710.
• the figure outlining operation 712 is conducted without user intervention, that is, the figure outlining operation 712 finds the path automatically based on seed and target pixel points and the information map.
• the technique used to compute the minimum cost path from seed to target is a modified Intelligent Scissors.
• the modified Intelligent Scissors approach in the present invention is to define a search band based on the location of seed and target pixel points.
• the modified Intelligent Scissors is to be detailed later.
• FIG. 4 is a flowchart 400 illustrating the human figure contour outlining method according to the present invention. Two distinct tasks are involved in this method, locating a person and finding the contour of the located person.
• the first task in this method is, primarily, accomplished by using a face detection process 402 more fully described in commonly assigned, co-pending U.S. Patent Application Serial No. 10/211,011, titled “Method For Locating Faces In Digital Color Images” filed August 02, 2002, by Shoupu Chen et al., and which is incorporated herein by reference; eye detection processes 404 described in commonly assigned, co- pending U.S. Patent Publication No. 2002/0114495 Al, filed December 19, 2000, titled “Multi-Mode Digital Image Processing Method For Detecting Eyes," and which is incorporated herein by reference; and statistics of human body measurement in "The Measure of Man and Women” by Alvin R Tilley, John Wiley & Sons, INC. 2001.
• the second task is tackled by using the Intelligent Scissors introduced above, aided with the parameters deduced from the eye detection process 404 and parameters obtained from the people detection 406. Note that localization of persons in images can also be completed with other methods. Face Detection
• FIG. 2 is a flow chart 200 illustrating the face candidate finding method according to the present invention. Each of these tests discards non-face objects with high confidence and retains most faces. The idea is to maintain a high true-positive candidate finding rate in every cascaded test, while keeping a relatively low false-positive rate for individual tests. There are basically four cascaded tests shown in FIG. 2.
• a chromaticity test 202 discards, with high confidence, non-skin-color pixels for the input color image.
• Chromaticity test 202 is different from conventional skin color detection methods used in face detection. In the past, in using color properties for face detection, most skin color detection techniques employed carefully designed skin color models in order to achieve high skin detection accuracy with very low false positives. However, skin color models having a high degree of accuracy often tended to exclude skin colors falling outside of the skin color region of a majority population upon which the skin color models were built. The exclusion of non-majority skin colors, in turn, results in face detection failures.
• the chromaticity test 202 in this design instead, focuses on exclusion of non-skin- colors with high accuracy. For example, it discards saturated green, or saturated blue pixels (sets pixels to zero), and keeps pixels having colors close to skin- colors. Therefore, it does retain most skin color pixels of non-majority populations.
• a geometry (or shape analysis) test 204 Pixels retained in the chromaticity test 202 are grouped into regions (or clusters of pixels). These regions are checked to see if they pass a geometry test.
• the geometry test 204 basically, checks a region's geometry shape, size, and location. Only those regions that pass the geometry test 204 will be allowed to enter a subsequent statistical test, that is, a grid pattern test 206. All pixels in regions failing the geometry test 204 are set to zero.
• regions or clusters of pixels possibly containing faces still remain and are further checked to locate approximate positions of faces. Localization of face candidates is performed by the grid pattern test 206 with the help of a mean grid pattern element (MGPe) image.
• the mean grid pattern element (MGPe) image is obtained in a step of Training 210.
• the grid pattern test 206 simply conducts a similarity check by evaluating a correlation coefficient between the MGPe image and a grid pattern element (GPe) image converted from a sub-image that is cropped from the remaining regions obtained from the geometry test step. Sub-images that pass the grid pattern test 206 are marked as face candidates.
• All face candidates are subsequently checked in a location test 208. Face candidates having a large portion residing in zero valued regions that were set in the geometry test 204 operation are unmarked in the location test 208 operation. Also, because the grid pattern test 206 is performed on the image in a raster scan fashion, it may result in multiple face candidates positioned very close to each other for the same face. A merge process is then performed to combine closely spaced multiple face candidates into a single face candidate based on a distance measure. Complete details of the face localization algorithm currently used can be found in the co-pending U.S. Patent Application Serial No. 10/211,011. Face detection also can be accomplished using other methods. Eye Detection An eye detection process 404, depicted in FIG.
• the current eye detection 404 adopts a method of Bayesian Iris-Based Multi-Modal Eye Localization described in the co-pending U.S. Patent Publication No. 2002/0114495 Al .
• FIG. 3 the flowchart of operations of the eye localization or eye detection process 404 is depicted in FIG. 3.
• a cluster generated in an operation of Clustering Iris Color Pixels 302 is a non-empty set of iris color pixels with the property that any pixel within the cluster is also within a predefined distance to another pixel in the cluster.
• Clusters are checked in an operation of Validating clusters 304.
• a cluster may be invalid because, for example, it may contain too many iris color pixels or because of its geometric properties. Invalid iris pixel clusters are removed from further consideration. If the number of valid iris color clusters "n" is at least two in an operation of query 306, then the process branches to an operation of "Template-less Eye Detection” 308 to find a pair of eyes. In operation 308, geometric reasoning is employed to detect eyes based on the geometric relationship between the iris pixel clusters.
• the process ends at an end operation 312.
• the process goes to an operation of "Template matching eye detection using valid iris clusters" 316.
• the process goes to an operation of "Template matching eye detection using image pixels" 314.
• Geometry relationship validation is still needed to make a final decision.
• the purpose of eye detection process 404 is to use a distance between two eyes as a precise reference measurement to deduce sizes of other body parts based on the statistical information provided in Tilley's paper. Certainly, the eye detection operation can be replaced by other facial feature finding schemes. Intelligent Scissors
• the Intelligent Scissors system has an interface that lets a user specify two pixels in an image, namely a seed point and a target point along an object's boundary. Then the system will try to find a cost-minimized path from the seed point to the target point. To find such a path, the image is first modeled as a graph. Every pixel in the image is converted to a node in the graph. Every node is then connected to its eight neighbors by links. Each link is associated with a cost. The cost value is determined by a cost function that is usually related to edge features.
• Links to pixels with strong edge features are associated with a low cost, and links to pixels with weak edge features are associated with a high cost.
• the initial cost map contains the cost associated with each link between any two 8-neighbors in the image.
• An exemplary cost map 600 is shown in FIG. 6.
• cst(q, r) Laplacian zero-crossing Z(r) Gradient magnitude G(r) Gradient direction D(q, r)
• Equation 1 where / z ( «)is a function related to zero-crossing feature, f G (») is a function related to gradient feature, f D (*) is a function related to gradient direction feature, w z , w fl and w G are user defined weights.
• a dynamic-programming path search algorithm similar to Dijkstra's algorithm is used to search for the optimal path from the seed to every single pixel, including the target pixel, in the image. Specifically, each pixel is given a pointer to the next pixel along the path.
• a modified approach in the present invention is to define a search band based on the location ofthe seed and target pixels.
• This search band is very similar to the rubber-band recently reported in a publication that proposed an improved Intelligent Scissors graph search algorithm in "Rubberband: An Improved Graph Search Algorithm for Interactive Object Segmentation," by Huitao Luo et al., Proc. oflCIP, 2002.
• Input s is a seed point t is a target point
• L is a rank (cost) ordered list of active nodes e(q) indicates whether node q has been expanded T(q) returns a total cost from q to the seed point cst(q, r) returns a cost between two neighboring pixels q and r min(Z) returns and removes a node with the lowest cost among nodes in list L
• SetSearchBand(s,t) sets a search band based on the current seed and target points
• N(q) returns up to 8 neighbors of q within the boundary set by SetBoundary(s, t)
• AddtoL(r) adds r to L at a proper position based on a cost associated with r
• the function SetSearchBand(s,t) generates a rectangular region around the seed and target pixels.
• the modified algorithm calculates a minimum cost path from each pixel within the band to the seed, starting from the seed itself (zero cost). Once these short paths near the seed are established, pixels that are farther and farther away from the seed are added on to paths which are already known to be optimal. A minimum-cost unexpanded pixel (node q in the above algorithm summary) is always guaranteed to have an optimal path, so by always choosing to expand this pixel next, the algorithm is guaranteed to only find optimal paths. A minimum cost path from target t to the seed is stored in B(t) . It should be noted that no path can start at the seed and go all the way around the object, back to the seed. This is because such a path would necessarily have a higher cost than the path including only the seed, as costs are always positive.
• a flowchart 400 for automatic human figure contour outlining is depicted in FIG. 4.
• a face detection process 402 which is described in FIG. 2 is applied to an incoming image to find a face region.
• the parameters ofthe face region are then used in the eye detection process 404 to locate the center between two eyes.
• the current flowchart 400 is designed to detect people that are upright in photo images. It is understood that the system can be modified to detect people in different postures.
• a distance, e (504) shown in FIG. 5
• H e (502) shown in FIG. 5
• H 6 e (512) shown in FIG. 5.
• positions and sizes of different upper body parts can be estimated in operation of people detection 406 (FIG. 4) using the coefficients given above.
• the information about body part locations is then used in a seed and target point generator 410 to produce a set of seed-target pixel pairs. It is understood that in order to use the Intelligent Scissors to accurately outline the body contour, these seed and target pixel pairs must be on or very close to edges of body parts captured in an image.
• a cost map ofthe image is computed in a cost map generator operation 408 in FIG. 4.
• the cost map is then fed to operation seed/target pixel pairs generator 410.
• the next operation is to find M key human figure contour points with M search windows.
• the exemplary cost map 600 is shown in FIG. 6 where five exemplary search windows for locating seed and target pixel pairs are also displayed. Note that brighter pixels indicate lower cost.
• Locations of search windows can be determined using the information obtained in the operation of people detection 406. For example, using the known eye vertical position (from the eye detection operation) and the coefficient H 2 , the vertical position of a top of a head 601 can be computed. A horizontal position ofthe top of head 601 is at a middle position ofthe corresponding pair of eyes. A center of window W x (604) can be located at the top of head 601. A width of W x could be chosen as an exemplary value of 1 (pixel).
• the height of W x could be equal to, for example, the length ofthe distance, e (504), between the two eyes. Then there will be e number of cost map pixels in window W (604) of this example. Once the size and position for window W x (604) are determined, within the window, collect N pixels with lowest cost values among the e pixels. An exemplary value currently used for N is 4. An exemplary minimum value for e is 14. The average vertical and horizontal positions of these N pixels are used to determine the key human figure contour points, P (602), shown in FIG. 6. P (602) will be used as a seed or a target point. Similarly, the positions of windows W 2 (608) and W 3 (610) can be determined using the known vertical eye position, the middle position ofthe two eyes, the value e, and coefficients H 3 (508) and H 6 (512). Windows ⁇ ,(614) and
• W s (618) can be determined using the known vertical eye position, the middle position ofthe two eyes, the value e, and coefficients H 3 (508), H 5 (510) and
• ⁇ (618) is e (pixels).
• W 4 (614), and W 5 (618) is one (pixel).
• any two of them can be grouped to form a seed and target pixel pair.
• an exemplary set of pairs could be the following: ⁇ (J?,/ > 2 );
• the first element in the above pairs is the seed point
• the second element is the target point
• the cost map generated in the operation of Cost Map Generator 408 is also fed into the Human Figure Outlining operation 412 to compute cst(q, r) , then to find the body contour using the set of seed and target points determined above.

Landscapes

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

Priority Applications (2)

Applications Claiming Priority (2)

Publications (1)

Family

ID=33298683

Family Applications (1)

Country Status (4)

Families Citing this family (34)

Citations (2)

Family Cites Families (4)

• 2003
  • 2003-04-23 US US10/421,428 patent/US7324693B2/en not_active Expired - Fee Related
• 2004
  • 2004-04-06 WO PCT/US2004/010529 patent/WO2004095358A1/en not_active Ceased
  • 2004-04-06 EP EP04759800A patent/EP1631933A1/en not_active Withdrawn
  • 2004-04-06 JP JP2006509730A patent/JP2006524394A/ja not_active Withdrawn

Patent Citations (2)

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events