EP1104570A1 - Procede de reconnaissance d'objets dans des images numerisees - Google Patents

Procede de reconnaissance d'objets dans des images numerisees

Info

Publication number
EP1104570A1
EP1104570A1 EP99945979A EP99945979A EP1104570A1 EP 1104570 A1 EP1104570 A1 EP 1104570A1 EP 99945979 A EP99945979 A EP 99945979A EP 99945979 A EP99945979 A EP 99945979A EP 1104570 A1 EP1104570 A1 EP 1104570A1
Authority
EP
European Patent Office
Prior art keywords
graph
bundle
jets
jet
image
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.)
Withdrawn
Application number
EP99945979A
Other languages
German (de)
English (en)
Inventor
Christian Eckes
Efthimia Kefalea
Christoph Von Der Malsburg
Michael Pötzsch
Michael Rinne
Jan C. Vorbrüggen
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Individual
Original Assignee
Individual
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 Individual filed Critical Individual
Publication of EP1104570A1 publication Critical patent/EP1104570A1/fr
Withdrawn legal-status Critical Current

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
    • G06V40/172—Classification, e.g. identification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR 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
    • G06V10/42—Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation
    • G06V10/422—Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation for representing the structure of the pattern or shape of an object therefor
    • G06V10/426—Graphical representations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
    • G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
    • G06V10/754—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries involving a deformation of the sample pattern or of the reference pattern; Elastic matching
    • 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
    • G06V40/20—Movements or behaviour, e.g. gesture recognition
    • G06V40/28—Recognition of hand or arm movements, e.g. recognition of deaf sign language

Definitions

  • the invention relates to a method for the automated recognition of one or more structures or one or more objects in digitized image data.
  • a method for facial recognition is known from DE 44 06 020.
  • so-called jets are extracted from a digitized image with gabor filters of different sizes and orientations, which are arranged at the nodes of a displaceable, scalable and deformable grid.
  • This graph i.e. the structure of the lattice and the jets associated with the nodes of the lattice are compared to a reference graph that includes the structure to be recognized.
  • the optimal shape of the grid is determined by a two-stage optimization of a graph comparison function. In the first phase, the size and position of the graph are optimized simultaneously; in the second phase, the intrinsic shape of the graph is optimized.
  • Another method is known from the field of face recognition, in which the comparison between the image of a head recorded with a video camera and several images of heads stored in a database is realized by a flexible imaging mechanism, the best possible image being determined by an optimization method (see Lades et al., IEEE Transactions on Computers, 42, 300-311).
  • a disadvantage of this method is that the method does not appear to be suitable for processing large amounts of data.
  • Lades et al. recognize an image of a head from a database consisting of images of 87 people; in a large number of applications, however, significantly larger reference databases can be expected.
  • the object of the invention is to improve the known methods in such a way that their robustness compared to less optimal image data is increased in comparison with the known method, wherein it should additionally be ensured that the method is compatible with conventional ones Funds can be realized.
  • This task is solved by a method for the automated recognition of one or more structures in digitized image data with the steps:
  • each reference graph having a network-like structure, which is defined in each case by assigning certain reference image data to nodes which are linked to one another in a predetermined manner and comprising jets, wherein a jet is assigned to each node and each jet comprises at least one partial jet which by folding at least one class of filter functions with different sizes and / or orientations with the reference image data of the corresponding reference image at the specific node or by folding at least one class of filter functions with different Sizes and / or orientations with color-segmented reference image data of the corresponding reference image at the specific node or by color information about the reference image data at the specific node or by means of statistical methods the texture descriptions of the reference image data obtained at the particular node or by motion vectors extracted from temporally successive reference images at the particular node,
  • a plurality of reference graphs can furthermore be made available, and those reference graphs which have network-like structures which are topologically identical, ie only by themselves distinguishing the lengths of corresponding links can be combined into a reference bundle graph.
  • a reference bundle graph in this case comprises a network-like structure which is defined by nodes which correspond to the nodes of the reference graphs and by links which are determined by averaging the corresponding links of the reference graphs, and bundle jets, each bundle being made up of the subjets which constitute the Jets correspond to the respective nodes which correspond to the reference graphs combined in the reference bundle graph.
  • an optimal image graph for the or each reference bundle graph is determined.
  • the optimal image graph for a specific reference bundle graph represents the optimal adaptation to this and is determined by projecting the network-like structure of the specific reference bundle graph into the image data, thereby defining the structure of the image graph, and determining sub-jets that at least a part of the Correspond to sub-jets that were used to determine the sub-jets of the reference graphs on which the specific reference bundle graph is based. Furthermore, the projection of the network-like structure of the specific reference bundle graph is varied until a graph comparison function, which compares the jets of the image graph with the corresponding bundle jets of the specific reference bundle graph, becomes optimal, partial jets of the image graph being compared with partial jets in the corresponding bundle jet of the specific reference bundle graph becomes. Finally, each structure is assigned to the reference image that corresponds to the reference graph or the reference graph from the reference bundle graph or graphs for which the graph comparison function is optimal in relation to the optimal image graph determined for it.
  • reference bundle graphs By using reference bundle graphs, the number of structures available for comparison can be increased with the same number of reference images, or in other words, a reference bundle graph allows a complex structure object class to be represented with a few examples. animals.
  • reference bundle graphs can be used to model such structure object classes using examples from individuals.
  • reference graphs can be combined to form one or more reference bundle graphs. This allows special cases within an object class to be treated separately or not taken into account.
  • all the reference graphs provided can be combined into one or more reference bundle graphs.
  • the methods described above can be developed in such a way that the structure of the jets assigned to the nodes, which is determined by the sub-jets, depends on the respective node.
  • Edge filters can also be used on the structure edge, for example.
  • the structure of the jets assigned to the nodes which is determined by the sub-jets, can be identical for all nodes.
  • a graph comparison function can advantageously be used, which includes a jet comparison function which takes into account the similarity of the jets corresponding to one another.
  • the graph comparison function can include a comparison function for the network-like structure, which takes into account the metric similarity of the image graph with the corresponding reference graph or the corresponding reference bundle graph.
  • the graph comparison function is expediently defined as the weighted sum of the jet comparison function and the comparison function for the network-like structure.
  • the jet comparison function can be defined as a function of individual jet comparison functions of corresponding jets.
  • the jet comparison function can advantageously be defined as a weighted sum of the individual jet comparison functions and / or as a weighted product from the individual jet comparison functions.
  • partial jets of the corresponding jets can expediently be taken into account and a single jet comparison function can be defined as a function of partial jet comparison functions.
  • the individual jet comparison function can advantageously be a weighted sum of the partial jet comparison functions and / or a weighted product of the partial jet comparison functions.
  • the bundle jets of the reference bundle graph B ⁇ can be divided into sub-bundle jets], and the jet comparison function between the sub-bundle jets ⁇ of the reference bundle graph and the corresponding sub-jets j n 'of the image graph G' for n nodes for m recursions can be calculated according to the following formulas :
  • ⁇ n is a weighting factor for the nth node n and the comparison function S n (B ", J n ') for the nth node of the reference bundle graph with the nth node of the image graph is given by:
  • ⁇ (M) max ( ⁇ , ⁇ [ 1) (M ⁇ 1) )], or
  • the sub-bundle jets of the reference bundle graph or graphs can only contain features which are folded by folding at least one class of filter functions with different sizes and / or orientations with the reference image data of the corresponding reference image at the specific node or by folding at least one class of filter functions with different sizes and / or orientations with color-segmented reference image data of the corresponding reference image at the particular node or by color information about the reference image data at the particular node or through texture descriptions of the corresponding reference image at the particular node obtained with statistical methods or by motion vectors at the particular node extracted from temporally successive reference images have been determined.
  • the sub-bundle jets of the reference bundle graph or graphs can only contain features that result from a reference graph.
  • sub-bundle jets of the one or more reference bundles can also comprise mixtures of these two aforementioned features.
  • a step for determining the significance of the recognition can be provided after the recognition of each structure.
  • an estimator can be used, for example, which takes into account both the optimal graph comparison function and the non-optimal graph comparison functions.
  • An estimator is particularly distinguished in this case, in which the distance between the values of the non-optimal graph comparison functions and the value of the optimal graph comparison function is determined.
  • these measures also provide information about the quality of the structure recognition.
  • each structure can be assigned to the reference images that correspond to the reference graphs or the reference graphs from the reference bundle graphs for which the values of the graph comparison functions lie in a predetermined range. If the values are not in the predetermined range, this means that a structure cannot be identified sufficiently. Accordingly, This further training is suitable for applications in which decisions are to be made based on the recognition process, such as for access control.
  • the color information can advantageously include color tone values and / or color saturation values and / or brightness values determined from the reference image data or the image data.
  • the reference graphs or the reference bundle graphs can be recalculated before each application, which is expedient in applications in which the reference data frequently change, in particular are updated, it is expedient in most applications that the step of making the reference graphs available or the reference bundle graph includes the retrieval of the reference graphs or the reference bundle graphs from a central database and / or a decentralized database, such as from chip cards.
  • the network-like structure of the reference graph in the form of a regular grid can be used, the nodes and links of which form right-angled meshes.
  • an irregular grid can be used as the network-like structure of the reference graph, the nodes and links of which are adapted to the structure to be recognized.
  • the nodes can be assigned characteristic points, so-called landmarks, to the structure to be recognized.
  • the jets are therefore determined at the characteristic points of the structure.
  • the characteristic see points when comparing the image data with the reference data is taken into account, whereby the significance with which a structure is recognized can be increased.
  • Gabor filter functions and / or Mallat filter functions can preferably be used as a class of filter functions for convolution with the reference image data or image data and / or as a class of filter functions for convolution with the color-segmented reference image data or image data.
  • the projection of the network-like structure of the specific reference graph or the specific reference bundle graph can comprise centering the reference graph or the specific reference bundle graph in the image.
  • the projection of the network-like structure of the specific reference graph or the specific reference bundle graph comprises a shift and / or a rotation of the centered reference graph or the centered reference bundle graph.
  • the projection of the network-like structure of the specific reference graph or the specific reference bundle graph can comprise scaling the centered reference graph or the centered reference bundle graph.
  • the significance and speed of recognition can be increased in particular if the structure to be recognized has a different size in the image data and the reference data.
  • the shift and / or the rotation and the scaling of the centered reference graph or of the centered reference bundle graph can be carried out simultaneously, whereby the recognition of a structure can be accelerated.
  • the projection of the network-like structure can include local distortions of the centered reference graph. This version is particularly suitable if image data and reference data have been recorded at different recording angles.
  • Such a local distortion can expediently be brought about by locally displacing a corresponding node of the centered reference graph.
  • the displacement or the scaling or the rotation can advantageously be determined from the comparison of the image graph with the corresponding reference graph or the corresponding reference bundle graph. This leads to a significant increase in the speed of recognition.
  • FIG. 1 shows an illustration of a hand position with a reference graph to explain an embodiment of the invention
  • Figure 2 shows the reference graph of Figure 1;
  • FIG. 3 shows a schematic illustration for determining a graph from a reference image according to an embodiment of the invention
  • FIG. 6 shows examples of images of a manual position with different backgrounds, which was compared with the images of FIG. 4 by means of an embodiment of the method according to the invention.
  • the various embodiments of the invention are described below using a method for recognizing the position of a hand, hereinafter referred to as the hand position.
  • this description is not to be understood as a restriction, but only as an example of a method for recognizing structures or objects in images.
  • FIG. 1 shows an example 20 of a hand in a certain position.
  • Figure 20 must be in digitized form.
  • the digitized form can either result directly from the recording process used, for example in Using a CCD camera, or must be digitized by converting an analog image, such as a conventional photograph.
  • a digitized image is typically in the form of a pixel field of a predetermined size.
  • Each pixel is assigned a horizontal and a vertical position x. Therefore, the pixel x is understood below to mean the pixel to which the position x has been assigned. Furthermore, each pixel is assigned a gray value and / or color information value, such as an HSI value.
  • the predetermined pixels are obtained by first projecting a net-shaped structure 21, which will be referred to as a grid for the sake of simplicity, into the image 20.
  • the grid used in the first embodiment comprises 15 nodes 22a, 22b, 22c, ... and 19 links, i.e. Connections between two nodes.
  • the links between the nodes 22a and 22b or 22b and 22c are provided with the reference symbols 23ab and 23bc.
  • the predetermined pixels are obtained by determining the pixels that correspond to the projection of the nodes.
  • an object-adapted grid is used, ie the nodes are assigned to characteristic points in the figure.
  • the nodes are therefore allocated to the two fingers and the back of the hand.
  • the invention is not limited to such object-adapted grids. Rather, grids in the actual sense, i.e. regular grids, can also be used. The decision as to which grid shape is appropriate depends on the respective application. Object-adapted lattice shapes generally lead to more significant recognition than regular lattices, but are more complicated to handle.
  • FIG. 3 shows a schematic illustration which explains how a representation in graph form can be obtained from FIG. 20, which, as will be explained, can be compared with other graphs, in particular with reference graphs.
  • two different classes 28 and 29 of features are used to compare the image with a reference image.
  • the first class of features 28a, ..., 28i is defined as the result of convolving the image on a predetermined pixel with a given filter function.
  • so-called complex Gabor filters are used as filter functions. These filters can be represented by the following formula:
  • the size and the orientation of the filter function can be determined by choosing the wave vector k.
  • Fig. 3 (c) the real parts of two such different filter functions are shown with 24a and 24b.
  • the value 0 is represented by a medium gray; positive values are lighter and negative values are darker.
  • the filter function 24a shown in FIG. 3 (c) has a low frequency or a small wave vector k, with an orientation of approximately 60 degrees with respect to the horizontal.
  • the filter function 24b has a larger frequency or a larger wave vector k, with an orientation of approximately 90 degrees with respect to the horizontal.
  • the feature J ⁇ x) can be calculated at a predetermined pixel x by:
  • a part of these features is shown schematically in the subjet 28.
  • the features 28a, 28b and 28c have been obtained by a filter function with a constant size and, as shown by the hatching, with different orientations.
  • features 28d, 28e and 28f as well as 28g, 28h and 28i result from a convolution with a smaller filter function.
  • any other filter functions can also be used.
  • so-called Mallat filters are used in accordance with a further embodiment of the invention.
  • the second class of features 29a, ..., 29i is defined as a result of convolution on a predetermined pixel of data resulting from conventional color space segmentation of the image data with a given filter function.
  • the color space segmentation was carried out with respect to the skin color.
  • the color space segmentation in Depending on the application, other colors can be performed.
  • Fig. 3 (b) the image data 22 is shown after the color space segmentation.
  • the convolution of the color space-segmented data was carried out analogously to the convolution of the image data, i.e. carried out with the same Gabor filters of the same size and orientation. A detailed description of this folding operation is therefore unnecessary and in this connection reference is made to the corresponding sections for folding the image data with the Gabor filters.
  • FIG. 3 (e) the color space segmented image data folded with the filter functions is shown in FIG. 3 (e).
  • Figures 25a and 25b in Fig. 3 (e) represent the real parts of a convolution of the color segmented image data with the filters, the real parts of which are designated in Fig. 3 (c) with 24a and 24b.
  • the sub-jets 28 and 29 form the jet 27 in the first embodiment.
  • the jet 27 finally comprises 80 features in this embodiment.
  • the convolution of the filter functions with the color-segmented data is not used as the second class, but rather color information obtained from the image data, such as hue, color saturation and intensity (HSI) for the pixel.
  • HSI hue, color saturation and intensity
  • a single jet is therefore composed of 40 features which result from the above-described convolution of the image data with the gabor filter functions and from the HSI triple and thus comprises 43 features.
  • a weighted Euclidean distance in the HSI color space is expediently used as a comparison function between two HSI triples.
  • a jet is composed of three classes.
  • the first class includes the features that result from the convolution of the image data with the Gabor filters (Gabor features); the second class includes the features that result from the convolution of the color space segmented image data with the Gabor filters (color lab features); and the third class comprises the HSI triples (color information) obtained from the image data.
  • a texture description obtained by statistical methods can be used as an alternative or in addition to the classes described above.
  • Such texture descriptions are For example, the mean gray value, the variance of the gray value distribution, the co-variance matrix of the gray value distribution, the entropy, the orientation of the gray value structure, mean scales of the structure in different directions, the range of variation of the local orientation, the range of variation of the spatial scales, the sequence and the arrangement different structural elements.
  • motion vectors are defined as a class in a further embodiment.
  • Such motion vectors can be calculated from two successive images using differential geometric methods (differential methods), correlation methods and filter methods.
  • differential geometric methods differential methods
  • correlation methods correlation methods
  • filter methods filter methods
  • a reference map must be provided in the form of a graph, ie in the form of a reference graph. This can be done, for example, by calling up the reference graph from a central database or a decentralized database, for example from chip cards.
  • this reference graph G is determined using one of the methods described above.
  • a grid as shown for example in FIG. 2, is accordingly projected into the reference image.
  • the jets are determined at the nodes of the grid.
  • the jets of the reference graphs should contain at least the feature classes that should also be determined for the comparison graph.
  • one projects the grid of the reference image (reference grid) into the comparison image and calculates the jets corresponding to this grid for the comparison image.
  • reference grid reference grid
  • the jets calculated in this way together with the structure of the projected grid finally form the comparison graph G '.
  • Different images can be used to project the reference grid into the comparison image, depending on the expected differences between reference images and comparison images.
  • the simplest projection namely centering the reference grid in the comparison image, is suitable, for example, if the reference image and comparison image have the same size and position with respect to the image center and were recorded at the same angle.
  • the following projections can optionally be carried out for this simple projection.
  • This mapping can be provided if the sizes of the reference mapping and the comparison mapping are different.
  • S S Jet + ⁇ S metric ⁇ (6)
  • S Jet denotes the jet comparison function, ie a suitable function that evaluates the similarity of the jets at corresponding points on the two graphs
  • S Metr ⁇ k denotes a suitable function that compares the similarity of the metrics of the two grids with one another.
  • S M ⁇ t ⁇ k strongly depends on the projection used.
  • ⁇ ( ⁇ > 0) denotes the weighting of the two comparison functions to each other, ⁇ can also be set to zero; this means that the similarity of the graph metrics is not taken into account. This value is particularly useful if only centering or displacement is selected as the projection, or in other words if the topology of the reference graph and the comparison graph is identical.
  • the comparison functions of the subjets for the respective classes k must first be calculated.
  • the features of which result from convolution of the image data and the image data segmented in the color space with Gabor filters, k would be 2.
  • a sub-jet comparison function is used, ie a function which depends on the amplitudes a and a,' of the two Depends on subjets and has the following form:
  • phase-sensitive comparison function can also be used, for example with the following form:
  • k j is the wave vector of the corresponding Gabor filters and d is an estimated displacement vector that compensates for rapid phase shifts, d is determined by the fact that S ⁇ in its Taylor expansion is internal
  • a single jet comparison function for corresponding individual jets can be formed from the partial jet comparison functions for the corresponding partial jets, that is to say for the characteristics of a class:
  • S k (j .j k ') are the partial jet comparison functions that can be calculated according to equations (7) or (8); ⁇ k are weighting coefficients that describe the contribution of a single subjet to the single jet comparison function S (j, j ').
  • the jet comparison function S Jet for all corresponding jets of a reference graph G and a comparison graph G ′ is formed from the individual jet comparison functions for the individual jets.
  • Various functions can also be used for this, depending on the application.
  • the comparison function for n jets can be formed according to:
  • S n (J n , J n ') are the individual jet comparison functions which can be calculated according to equations (10) to (13); ⁇ are weighting coefficients, which describe the contribution of a single jet to the jet comparison function S Jel .
  • the selection of the suitable comparison function for a given structure of the comparison images and the reference images can be determined by comparison tests with the comparison images and the reference images and is therefore within the range of the average skill of a person skilled in the art.
  • the comparison graph can now be optimally adapted to the reference graph.
  • the projection of the reference grid in the comparison image is varied until the graph comparison function assumes an optimal value (in the case of the comparison functions described above, this is a minimum).
  • the comparison image can also be used to compare several reference images.
  • the optimal adaptation to each reference image and the graph comparison function for this optimal adaptation are determined for the comparison image using the method described above.
  • the reference map which has the greatest similarity to the comparison map can be determined.
  • a measure of the significance of the similarity can be determined for all reference images by evaluating the graph comparison functions. Depending on the degree of similarity required, different definitions can be used for a significant recognition. For example, the mean value S and the variance ⁇ s can be formed from all the graph comparison functions for the non-optimal reference graphs. A significant similarity could be assumed if
  • This embodiment differs from the previously described embodiments in the structure of the reference graphs, which are compared with the comparison graph.
  • a reference graph was created from a single reference image
  • a reference bundle graph results from several reference images.
  • M model graphs are created from M images, which are summarized in a so-called reference bundle graph.
  • All M model graphs have the same qualitative structure, that is to say they each have N nodes which are connected to one another by a predetermined grid. However, it is permissible here that the lengths of two correspond to one another. the links are different. Only the topological identity of the model graph is required.
  • the structure of the lattice can have a regular shape, as shown here, or an irregular shape.
  • the grids used in the previously discussed embodiments that is to say a regular nxm grating or an irregular object-adapted grating, can accordingly be used.
  • the corresponding jets of the M model graphs also differ.
  • the M model graphs are finally combined into a bundle graph, as explained below.
  • the mean distance vectors ⁇ - tl between nodes i and j in the bundle graph are determined by:
  • ⁇ TM is the distance vector between nodes i and j in the model graph m.
  • Bundle jets of the M model graphs are also assigned to the nodes of the bundle graph. These bundle jets include all sub-jets that were determined when the reference graphs were created. Accordingly, a jet of the bundle graph comprises the M corresponding jets of the M model graphs. In other words, a jet of the reference bundle graph is constructed from the sub-jets, which each correspond to the sub-jets of the model graphs.
  • the bundle jets are assigned to sub-bundle jets.
  • the characteristic classes, the reference images or mixed forms of the two can be used, for example, as an assignment criterion.
  • the sub-bundle jets When assigned to feature classes, the sub-bundle jets only comprise the features of a single feature class.
  • the jet comparison function S jet is determined here, as will be explained in the following.
  • the comparisons of the partial jets with the partial bundle jets are functionally related to the comparison function for the Jet J with the bundle graph B M :
  • the function ⁇ can be determined using the following recursion:
  • ⁇ (0) (M) ⁇ ⁇ , ⁇ [ 1) (M [ 1) ), or (25)
  • ⁇ (0) (M) ⁇ ( ⁇ [ 1) (M [ 1) )) ⁇ ' , or (26)
  • ⁇ (0) (M) max ⁇ , ⁇ ⁇ 1) (M 1) ) ⁇ , or (27)
  • the jet comparison function S Jet for all corresponding jets of a reference bundle graph B M and a comparison graph G ' is finally formed.
  • Various functions can also be used for this, depending on the application.
  • the comparison function for n nodes can be formed according to:
  • This comparison function can finally be used in equation (6) to determine the graph comparison function.
  • the jets with the maximum similarity are selected from the bundle graph.
  • a bundle graph can be treated like a single graph, it is also possible to combine several reference bundle graphs, one reference bundle graph and one or more reference graphs in a database with which the comparison graphs are to be compared.
  • Each reference graph has 15 nodes and 20 links. The nodes were manually distributed at anatomically significant points.
  • FIG. 6 shows various examples of a hand position in front of different backgrounds. A total of 29 backgrounds were used, five of which had a high level of skin tone, eleven had a medium level of skin tone and eight had a low level of skin tone.
  • the first reference bundle graph contained all Gabor features
  • the second reference bundle graph all Color Laboratory features
  • the third reference bundle graph all HSI features.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Multimedia (AREA)
  • General Physics & Mathematics (AREA)
  • Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Human Computer Interaction (AREA)
  • Software Systems (AREA)
  • Medical Informatics (AREA)
  • Evolutionary Computation (AREA)
  • Databases & Information Systems (AREA)
  • Computing Systems (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Artificial Intelligence (AREA)
  • Psychiatry (AREA)
  • Social Psychology (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)

Abstract

Le procédé selon l'invention comprend les étapes suivantes: (a) mise à disposition d'au moins un graphe de référence, constitué d'images de référence, le ou chaque graphe de référence présentant une structure du type réseau qui est chaque fois définie par le fait que des noeuds, qui sont reliés entre eux par des liaisons d'une façon prédéterminée, sont attribués à certaines données d'image de référence, ainsi que des jets, un jet étant affecté à chaque noeud et chaque jet comprenant au moins un jet partiel qui est déterminé par convolutions d'au moins une classe de fonctions de filtrage, avec différentes grandeurs et/ou orientations, au moyen des données d'image de référence, ou par convolutions d'au moins une classe de fonctions de filtrage avec différentes grandeurs et/ou orientations, à l'aide de données d'image de référence segmentées par couleur, ou bien par des informations de couleur concernant les données d'image de référence, ou bien par écriture d'une texture de l'image de référence correspondante, ou bien par des vecteurs de déplacement au niveau des noeuds déterminés; (b) détermination d'un graphe d'image optimal à partir des données d'image numérisées pour chaque graphe de référence, le graphe d'image optimal représentant, pour un graphe de référence déterminé, l'adaptation optimale à celui-ci; et (c) attribution de la ou de chaque structure à l'image de référence qui correspond au graphe de référence pour lequel la fonction de comparaison graphique est optimale par rapport au graphe d'image optimal déterminé pour ledit graphe de référence.
EP99945979A 1998-08-14 1999-08-13 Procede de reconnaissance d'objets dans des images numerisees Withdrawn EP1104570A1 (fr)

Applications Claiming Priority (3)

Application Number Priority Date Filing Date Title
DE19837004 1998-08-14
DE19837004A DE19837004C1 (de) 1998-08-14 1998-08-14 Verfahren zum Erkennen von Objekten in digitalisierten Abbildungen
PCT/EP1999/005946 WO2000010119A1 (fr) 1998-08-14 1999-08-13 Procede de reconnaissance d'objets dans des images numerisees

Publications (1)

Publication Number Publication Date
EP1104570A1 true EP1104570A1 (fr) 2001-06-06

Family

ID=7877611

Family Applications (1)

Application Number Title Priority Date Filing Date
EP99945979A Withdrawn EP1104570A1 (fr) 1998-08-14 1999-08-13 Procede de reconnaissance d'objets dans des images numerisees

Country Status (5)

Country Link
US (1) US7113641B1 (fr)
EP (1) EP1104570A1 (fr)
AU (1) AU5851999A (fr)
DE (1) DE19837004C1 (fr)
WO (1) WO2000010119A1 (fr)

Families Citing this family (53)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE10022570A1 (de) * 2000-05-09 2001-11-15 Giesecke & Devrient Gmbh Verfahren und System zur Generierung eines Schlüsseldatensatzes
DE50102061D1 (de) * 2001-08-01 2004-05-27 Zn Vision Technologies Ag Hierarchische Bildmodellanpassung
GB2384095B (en) * 2001-12-10 2004-04-28 Cybula Ltd Image recognition
US8494286B2 (en) 2008-02-05 2013-07-23 DigitalOptics Corporation Europe Limited Face detection in mid-shot digital images
US8363951B2 (en) 2007-03-05 2013-01-29 DigitalOptics Corporation Europe Limited Face recognition training method and apparatus
US9692964B2 (en) 2003-06-26 2017-06-27 Fotonation Limited Modification of post-viewing parameters for digital images using image region or feature information
US7792335B2 (en) * 2006-02-24 2010-09-07 Fotonation Vision Limited Method and apparatus for selective disqualification of digital images
US7269292B2 (en) 2003-06-26 2007-09-11 Fotonation Vision Limited Digital image adjustable compression and resolution using face detection information
US7792970B2 (en) 2005-06-17 2010-09-07 Fotonation Vision Limited Method for establishing a paired connection between media devices
US8155397B2 (en) 2007-09-26 2012-04-10 DigitalOptics Corporation Europe Limited Face tracking in a camera processor
US8682097B2 (en) 2006-02-14 2014-03-25 DigitalOptics Corporation Europe Limited Digital image enhancement with reference images
US7471846B2 (en) 2003-06-26 2008-12-30 Fotonation Vision Limited Perfecting the effect of flash within an image acquisition devices using face detection
US8330831B2 (en) 2003-08-05 2012-12-11 DigitalOptics Corporation Europe Limited Method of gathering visual meta data using a reference image
US7574016B2 (en) 2003-06-26 2009-08-11 Fotonation Vision Limited Digital image processing using face detection information
US8896725B2 (en) 2007-06-21 2014-11-25 Fotonation Limited Image capture device with contemporaneous reference image capture mechanism
US8553949B2 (en) 2004-01-22 2013-10-08 DigitalOptics Corporation Europe Limited Classification and organization of consumer digital images using workflow, and face detection and recognition
US8593542B2 (en) 2005-12-27 2013-11-26 DigitalOptics Corporation Europe Limited Foreground/background separation using reference images
US9129381B2 (en) 2003-06-26 2015-09-08 Fotonation Limited Modification of post-viewing parameters for digital images using image region or feature information
US7620218B2 (en) 2006-08-11 2009-11-17 Fotonation Ireland Limited Real-time face tracking with reference images
US8989453B2 (en) 2003-06-26 2015-03-24 Fotonation Limited Digital image processing using face detection information
US7565030B2 (en) 2003-06-26 2009-07-21 Fotonation Vision Limited Detecting orientation of digital images using face detection information
US8498452B2 (en) 2003-06-26 2013-07-30 DigitalOptics Corporation Europe Limited Digital image processing using face detection information
US8948468B2 (en) 2003-06-26 2015-02-03 Fotonation Limited Modification of viewing parameters for digital images using face detection information
US7844076B2 (en) 2003-06-26 2010-11-30 Fotonation Vision Limited Digital image processing using face detection and skin tone information
US7440593B1 (en) * 2003-06-26 2008-10-21 Fotonation Vision Limited Method of improving orientation and color balance of digital images using face detection information
WO2005038700A1 (fr) * 2003-10-09 2005-04-28 University Of York Reconnaissance d'images
US7564994B1 (en) 2004-01-22 2009-07-21 Fotonation Vision Limited Classification system for consumer digital images using automatic workflow and face detection and recognition
US8320641B2 (en) 2004-10-28 2012-11-27 DigitalOptics Corporation Europe Limited Method and apparatus for red-eye detection using preview or other reference images
US7715597B2 (en) 2004-12-29 2010-05-11 Fotonation Ireland Limited Method and component for image recognition
US8488023B2 (en) * 2009-05-20 2013-07-16 DigitalOptics Corporation Europe Limited Identifying facial expressions in acquired digital images
US8503800B2 (en) 2007-03-05 2013-08-06 DigitalOptics Corporation Europe Limited Illumination detection using classifier chains
US7315631B1 (en) 2006-08-11 2008-01-01 Fotonation Vision Limited Real-time face tracking in a digital image acquisition device
US7804983B2 (en) 2006-02-24 2010-09-28 Fotonation Vision Limited Digital image acquisition control and correction method and apparatus
ATE497218T1 (de) 2006-06-12 2011-02-15 Tessera Tech Ireland Ltd Fortschritte bei der erweiterung der aam- techniken aus grauskalen- zu farbbildern
EP2050043A2 (fr) 2006-08-02 2009-04-22 Fotonation Vision Limited Reconnaissance faciale avec ensembles de données combinés basés sur une acp
US7403643B2 (en) 2006-08-11 2008-07-22 Fotonation Vision Limited Real-time face tracking in a digital image acquisition device
US7916897B2 (en) 2006-08-11 2011-03-29 Tessera Technologies Ireland Limited Face tracking for controlling imaging parameters
US8055067B2 (en) 2007-01-18 2011-11-08 DigitalOptics Corporation Europe Limited Color segmentation
DE602008001607D1 (de) * 2007-02-28 2010-08-05 Fotonation Vision Ltd Trennung der direktionalen beleuchtungsvariabilität in der statistischen gesichtsmodellierung auf basis von texturraumzerlegungen
WO2008109622A1 (fr) 2007-03-05 2008-09-12 Fotonation Vision Limited Catégorisation de visage et annotation d'une liste de contacts de téléphone mobile
JP4970557B2 (ja) 2007-03-05 2012-07-11 デジタルオプティックス・コーポレイション・ヨーロッパ・リミテッド デジタル画像取込装置における顔検索及び検出
US7916971B2 (en) 2007-05-24 2011-03-29 Tessera Technologies Ireland Limited Image processing method and apparatus
US8750578B2 (en) * 2008-01-29 2014-06-10 DigitalOptics Corporation Europe Limited Detecting facial expressions in digital images
US20090200467A1 (en) * 2008-02-12 2009-08-13 Gray Paul C Automatic image-based volumetric detection of an object in a space
US7855737B2 (en) 2008-03-26 2010-12-21 Fotonation Ireland Limited Method of making a digital camera image of a scene including the camera user
CN106919911A (zh) 2008-07-30 2017-07-04 快图有限公司 使用脸部检测的自动脸部和皮肤修饰
WO2010063463A2 (fr) 2008-12-05 2010-06-10 Fotonation Ireland Limited Reconnaissance de visage au moyen de données de classificateur de suiveur de visage
US8379917B2 (en) 2009-10-02 2013-02-19 DigitalOptics Corporation Europe Limited Face recognition performance using additional image features
KR101633198B1 (ko) * 2011-12-28 2016-06-23 엠파이어 테크놀로지 디벨롭먼트 엘엘씨 객체 맥락적 정보의 분류 방지
US9036044B1 (en) * 2013-07-22 2015-05-19 Google Inc. Adjusting camera parameters associated with a plurality of images
CN105095923B (zh) * 2014-05-21 2018-09-11 华为技术有限公司 一种图像处理方法及装置
EP3564847A1 (fr) 2018-05-02 2019-11-06 IDEMIA Identity & Security Germany AG Identification d'objet
EP3570207B1 (fr) 2018-05-15 2023-08-16 IDEMIA Identity & Security Germany AG Témoins vidéo

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE4406020C1 (de) * 1994-02-24 1995-06-29 Zentrum Fuer Neuroinformatik G Verfahren zur automatisierten Erkennung von Objekten
US6222939B1 (en) * 1996-06-25 2001-04-24 Eyematic Interfaces, Inc. Labeled bunch graphs for image analysis
DE19726226C2 (de) * 1997-06-22 2001-07-26 Zentrum Fuer Neuroinformatik G Verfahren zum automatisierten Erkennen von Strukturen in Schnitten durch biologische Zellen oder biologisches Gewebe
KR100530812B1 (ko) * 1998-04-13 2005-11-28 네브엔지니어링 인코포레이티드 아바타 애니메이션을 위한 웨이브렛-기반 얼굴 움직임 캡쳐
US6301370B1 (en) * 1998-04-13 2001-10-09 Eyematic Interfaces, Inc. Face recognition from video images
US6714661B2 (en) * 1998-11-06 2004-03-30 Nevengineering, Inc. Method and system for customizing facial feature tracking using precise landmark finding on a neutral face image
US6580821B1 (en) * 2000-03-30 2003-06-17 Nec Corporation Method for computing the location and orientation of an object in three dimensional space

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See references of WO0010119A1 *

Also Published As

Publication number Publication date
DE19837004C1 (de) 2000-03-09
AU5851999A (en) 2000-03-06
US7113641B1 (en) 2006-09-26
WO2000010119A1 (fr) 2000-02-24

Similar Documents

Publication Publication Date Title
DE19837004C1 (de) Verfahren zum Erkennen von Objekten in digitalisierten Abbildungen
DE69324207T2 (de) Bildgruppierungsvorrichtung
DE69938173T2 (de) Automatische voreingestellte teilnehmerpositionsbestimmung für videokonferenzsysteme
DE69026329T2 (de) Entrasterung von gespeicherten digitalen Halbtonbildern durch logisches Filtrieren
DE69624758T2 (de) Bildsegmentierungssystem und -verfahren
DE4406020C1 (de) Verfahren zur automatisierten Erkennung von Objekten
EP1316057B1 (fr) Evaluation d'informations de direction d'aretes
DE19634768A1 (de) Vorrichtung und Verfahren zur Erfassung eines Gesichts in einem Videobild
DE69322095T2 (de) Verfahren und gerät zur identifizierung eines objekts mittels eine geordneten folge von grenz-pixel-parametern
WO1992008204A2 (fr) Procede pour la reconnaissance et l'evaluation de la position dans l'espace d'objets a partir d'une image bidimensionnelle
WO1995009403A1 (fr) Procede de segmentation d'images numeriques en couleur
DE102007013971A1 (de) Verfahren und Vorrichtung zur Ermittlung einer Zellkontur einer Zelle
DE19928231C2 (de) Verfahren und Vorrichtung zur Segmentierung einer Punkteverteilung
DE69612284T2 (de) Datenverarbeitungsverfahren- und vorrichtung
WO2008034599A2 (fr) Procédé et dispositif de traitement d'image
DE3923914A1 (de) Verfahren zum extrahieren eines zeichenmerkmals
DE69008637T2 (de) Iteratives Bewegungsabschätzungsverfahren zwischen einem Referenzbild und einem aktuellen Bild, und Verfahren zu ihrer Herstellung.
EP0394959A2 (fr) Méthode d'analyse d'image
EP1282070B1 (fr) Adaptation hierarchique d'un modèle d'un image
EP1525442B1 (fr) Procede pour analyser des divergences de couleur dans des images a l'aide d'un capteur d'image
DE69322216T2 (de) Verfahren und Gerät zur graphischen Mustererzeugung
DE19754909A1 (de) Verfahren und Vorrichtung zur Erfassung und Bearbeitung von Abbildungen biologischen Gewebes
DE19834718C2 (de) Digitale Bildverarbeitung für ein Qualitätskontrollsystem
DE19726226C2 (de) Verfahren zum automatisierten Erkennen von Strukturen in Schnitten durch biologische Zellen oder biologisches Gewebe
AT505885B1 (de) Verfahren zum ähnlichkeitsvergleich von zwei gegenständen

Legal Events

Date Code Title Description
PUAI Public reference made under article 153(3) epc to a published international application that has entered the european phase

Free format text: ORIGINAL CODE: 0009012

17P Request for examination filed

Effective date: 20010314

AK Designated contracting states

Kind code of ref document: A1

Designated state(s): AT BE CH CY DE DK ES FI FR GB GR IE IT LI LU MC NL PT SE

17Q First examination report despatched

Effective date: 20030915

REG Reference to a national code

Ref country code: DE

Ref legal event code: 8566

RBV Designated contracting states (corrected)

Designated state(s): FR GB

STAA Information on the status of an ep patent application or granted ep patent

Free format text: STATUS: THE APPLICATION HAS BEEN WITHDRAWN

18W Application withdrawn

Effective date: 20071009