SE535888C2 - Procedure for identifying a characteristic part of an image - Google Patents

Abstract

The invention is characterized in the steps: a) identifying an edge area (10) in the image; b) identifying a first direction (Al) in the image for the edge area (10); c) identifying a local, connected edge area (11) the width of which has one single global maximum; d) identifying a second direction (A2) for the local edge area (11); e) identifying a supplemental area (12) around the local edge area (11); f) identifying the characteristic part of the image so that it comprises the local edge area (11) and the supplemental area (12).

Description

5 535 888 several characteristic parts of a certain image are identified. Such a characteristic part can be identified in different ways, but with the common purpose of being information-bearing in the sense that the part can be used to, for example, identify or compare an image instead of using the entire image. If the characteristic part contains information that is representative or descriptive of the image as a whole, it is more effective in different image analyzes to take into account only the characteristic part rather than the whole image.

One problem is to be able to automatically identify relevant characteristic parts in images in a consistent and repeatable way.

WO2006085861 describes a method for identifying or indexing a digitally stored image, in which characteristic parts are identified as edge areas, defined as sets of pairs of pixels with high light contrast.

When performing an analysis based on a characteristic image part, for example by comparing it with a corresponding characteristic image part in another image, problems often arise when the image in which one characteristic part is included changes, for example by shifting perspective. , angle change, rotation, resolution change, cropping, color changes and so on. Examples include comparison between two images by means of respective characteristic parts where one image is a variant of the other image that has been cropped or rotated; the identification of deviating images in an image material by means of respective characteristic parts where certain images have deviating resolution but where the resolution itself is not the property whose deviation is what is meant by the identification; and grouping a large number of images according to motifs based on the respective characteristic parts, where certain images include the same motif * but shown from slightly different angles or with partly different backgrounds.

The common problem in these cases is to provide a method for identifying a characteristic image part in a consistent and repeatable manner, for the purposes described above, and which is as unaffected as possible by image changes of the types mentioned above.

The present invention solves the problems described above.

Thus, the invention relates to a method for identifying a characteristic part of an image, wherein the image is caused to be stored in digital format in the form of a matrix made up of pixels, and is characterized by the following steps: a) identifying an edge area in the image consisting of of a coherent amount of edge pixels the image; b) which together form an edge area in identifying a first direction in the image for the edge area; c) identify a local edge area consisting of a coherent number of pixels, for which local edge area the width in a first width direction along the first direction lacks local maximums except for a single global maximum; d) identify a second direction for the local border area; e) identifying an additional area around the local edge area that includes pixels located in close proximity to a peripherally located boundary pixel of the local edge area and further away from the boundary pixel up to a distance in the other direction from the boundary pixel proportional to the width of the local edge area; in the other direction at the boundary pixel: f) identifying the characteristic part of the image so as to include the local edge area and the additional area, and storing the characteristic part in digital form on a storage medium.

The invention will now be described in detail, with reference to exemplary embodiments of the invention and the accompanying drawings, in which: Figure 1 is an enlarged detail in a digitally stored image to which a method according to the present invention is applied; Figure 2 is a graph illustrating a threshold value for a comparison parameter according to the present invention; Figure 3 is a flow chart illustrating the identification of an edge area in accordance with the present invention; Figure 4 is a schematic illustration of an identification of a characteristic area according to the present invention; and Figure 5 is a flow chart illustrating the identification of a characteristic part in accordance with the present invention.

Figure 1 shows a greatly enlarged section 1 of an image.

In accordance with the invention, the image is stored in digital format on a digital storage medium in the form of a matrix made up of pixels. This means that the image can be an image whose model is an analogously stored image, for example stored on a photographic film. In this case, a conventional digitization of the image is performed before it is used in a method according to the present invention. The storage medium can be any conventional digital storage medium, such as a hard disk, an internal memory of one or more computers or a CD. BB 25 In Figure 1, the section is enlarged so that the individual pixels are clearly visible. A grid has been added, solely for the purpose of further increasing the clarity of the figure.

Figure 3 briefly illustrates the various steps performed in identifying an edge area in accordance with a preferred embodiment of the present invention.

In a first step 101, one sweeps over the entire image, pixel by pixel, and calculates for each pair of pixels in the image, where one pixel is located. A predetermined local environment to the other pixel, a comparison parameter that is the difference between the two pixels regarding an intensity property over one or more channels.

It is preferred that the intensity property difference be defined as the absolute difference in the value of the intensity property. For example, the difference may be the absolute difference in luminance of two pixels.

It is also preferred that a difference in intensity between the pixel in question and each of the pixels included in a predefined local environment be compared to the pixel. The environment is preferably symmetrical about the pixel in question, and may, depending on, among other things, performance requirements, 4, 8, for example consist of 12, 20, 24 or more adjacent pixels.

Figure 1 illustrates a pixel 4 for which the difference in intensity in comparison with each of the pixels 3a-3h in an 8 pixel 3x3 environment 3 (the middle pixel 4 is the comparison pixel and does not belong to the environment 3) is illustrated. 10 15 20 25 30 535 BBB Thus, for each pixel 4 in the image, an intensity difference between the pixel 4 and each of the pixels 3a-3h in an environment 3 is compared to the pixel 4. According to a preferred embodiment, each ambient pixel 3a 3h a direction in the image in relation to the pixel 4. Table 1 shows the different directions, expressed as angles in relation to the geometry of the image, to the pixel 4 which the different ambient pixels 3a ~ 3h define in figure 1. It is of course possible that use fewer or more directions, but it is preferred that the directions be evenly distributed between 0 and 360 °.

Table 1 Ambient Pixels Direction 3a 270 ° 3b 3l5 ° 3c 0 ° 3d 45 ° 3e 225 ° 3f l80 ° 3g l35 ° 3h 90 ° According to the presently preferred embodiment, for each pixel 4 and for each direction used, a comparison parameter is calculated, with respect to an intensity property. over one or more channels, which is the difference between the pixel 4 and the ambient pixel defining the direction in question. A "channel" in this context is a numerical indication of a certain type of intensity, such as luminance, reflectance or the like.

Examples include an RGB-coded image, where light intensity in the red, green and blue ribbons is indicated numerically by M7 535 BBB each of three channels, respectively, and a grayscale image, where grayscale is indicated numerically in a single channel.

According to a preferred embodiment, the intensity property of a given pixel is the total luminance of the pixel in question over the entire visible color spectrum. In case the original image contains color information, it is preferable that the image in an initial step, for example in connection with the above-described possible digitization, is converted to a grayscale image, in which the total luminance for a certain pixel consists of the value of its gratin. By ignoring color information in the image in this way, a characteristic part of the image can be identified in a way that is repeatable even in many cases where the color information of the original image has been manipulated.

For all pixels in the image, the calculated comparison value is stored for each of the surrounding pixels in the environment for each respective pixel. The amount of comparative values thus obtained can be illustrated for each of the directions used in the image in a graph of the type illustrated in Figure 2, which along one axis d shows difference in intensity property and along another axis n shows the number of pixels in the image that show this intensity property difference for the current direction. In other words, eight such graphs can be created for the sweep illustrated in Figure 1 using a 3x3 pixel environment.

For each graph, in a step 102, one is then identified which constitutes the first local minimum of the function n (d) which for a certain direction is illustrated in threshold value dl, figure 2. When calculating the value dl it is preferable to first perform filtering the function f (n) so that it becomes smoother and / or introducing a certain tolerance parameter that ignores minor fluctuations in the value of the function. This ensures that it is the first more prominent minimum that is selected as the threshold value dl.

In a step 103, an edge area in the image is then identified as a contiguous area of edge pixels. The expression "edge pixel" in this context is to be interpreted as the pixels for which the comparison parameter in comparison with another pixel in step 101 was found to be greater than the threshold value dl. In other words, edge pixels can be said to be the pixels in the image that show the highest contrast in comparison with their respective local environments. The same image may include any number of such contiguous areas, but it is preferred that each contiguous and isolated area be defined as a single edge area, possibly using certain thresholds in the form of, for example, the minimum allowable size for an edge area or the minimum allowable size. for connections between edge areas. Although it is preferable to perform the comparisons and determinations of threshold values as described above in each of certain predetermined directions in the image, it is also possible to perform the intensity property comparison between the pixel 4 and the surrounding pixels 3a-3h without taking into account directions. which in the case illustrated in Figure 1 would result in 8 different comparison values for each pixel in the image, and where all these comparison values are analyzed in one and the same function f (n), so that only one threshold value dl will be determined.

In other words, if directions are used, a particular pixel 4 will be considered an edge pixel if the intensity difference between the pixel 4 and any of its surrounding pixels 3a-3h is greater than the threshold value of the direction defined. of the ambient pixel in question. In the event that directions are not used, on the other hand, if the pixel 4 is considered to be an edge pixel intensity difference and between the pixel 4 one of the ambient pixels 3a-3h is greater than the threshold common to all directions.

According to a preferred embodiment, each edge pixel thus identified is associated with a binary number, in which each bit corresponds to a certain direction, and where the value of the respective bit (0 or 1) relates to whether the comparative parameter of the edge pixel in question in the direction corresponding to the bit was found to be greater than the threshold value for at least one of the predetermined directions. If, for example, the pixel 4 in Figure 1 was found to be an edge pixel because its difference in intensity in comparison with the surrounding pixels 3d, 3e and 3g was larger than the calculated threshold values for these three directions, the binary number can be calculated according to Table 2: Table 2 Ambient pixel Direction Bit value 3a 270 ° 0 3b 3l5 ° O 3c 0 ° 0 3d 45 ° l 3e 225 ° 1 3f l80 ° 0 3g l35 ° 1 3h 90 ° O 10 20 25 30 535 BBB 10 Thus the binary number 00011010, which corresponds to the decimal number 26. Such a representation, where a unique binary number identifies the directions along which the edge pixel 4 in question meets the criterion of being considered an edge pixel, allows efficient search among identified edge pixels because each edge pixel is associated with information. indicating in which directions adjacent pixels are at least likely also edge pixels.

According to the invention, an identified edge area is used, consisting of a continuous amount of edge pixels which together form an edge area in the image, fields, for example between two adjacent ones, to identify a characteristic area. It is preferred but not necessary to use a method in accordance with that described above to identify the edge area. By identifying as pixel pixels such pixels that have a large intensity property difference compared to other pixels in a local environment, an edge area is created which effectively captures relevant parts of the image and is in many ways independent of the types of changes to the original image described. above. By also performing the analysis in a number of predetermined directions, the precision is improved.

The flow chart in Figure 5 illustrates a method according to the present invention, where the identification of the edge area constitutes a first step 201.

Figure 4 shows in a matrix a certain edge area 10 identified in step 201 for a certain image. The image information itself is not displayed, for the sake of clarity, but only the extent of the edge area 10 in the pixel matrix of the image. The pixels belonging to 10 15 20 25 30 535 BBB ll the identified edge area 10 are marked with a gray background.

In accordance with the invention, in a step 202, a first direction A1 in the hand is identified. The first direction A1 can be any direction, and can in particular be a direction in relation to the axis directions of the image or in relation to a certain geometric property of the identified edge area, such as its longest diameter. The direction A1 illustrated in Figure 4 is one of the axis directions of the image.

In a step 203, the entire identified edge area 10 is then traversed along the first direction A1, in the exemplary embodiment illustrated in Figure 4 from below and upwards in the figure. For each pixel along the first direction A1, the width in a first width direction for the edge area 10 at the position of the pixel is examined. By "the width in a first width direction" is meant in this case. the term context width in an additional direction that is not parallel to the first direction A1. This width direction is preferably perpendicular to the first direction A1, but can also be constituted, for example, by an axis direction not parallel to A1 in the image.

By such successive investigations of the width in the first width direction along the first direction A1, a local edge area 11 is identified in a step 203. For each examined width in the first width direction, the local edge area ll is added to the pixels located along the width in the first width direction in question. According to the invention, the local edge area ll is identified so that the width in the first width direction in the local edge area ll along the first direction A1 lacks local maximums except for a single global maximum. This can be practically done by continuing the passage to the next width in the first width direction along the first direction A1 as long as the width in the first width direction does not start to increase again after first decreasing. In other words, the identification of the local edge area l1 is interrupted when the first local minimum of the width in the first width direction has been reached along the first direction A1, alternatively when the last width in the first width direction in the edge area 10 has been reached.

According to a preferred embodiment, the previously stored binary numbers for each respective edge pixel are used to perform the traversal of the edge area 10. This is done so that the edge area 10 traverses pixel by pixel, where the bit in the binary number corresponding to a certain direction determines whether the throughput should continue in this direction or not in such a way that the bit value “O” means that the throughput should not continue in that direction. Such a method results in rapid traversal of all the pixels in an edge area 10.

The local edge area 11 identified in this way is illustrated in Figure 4 by means of dashed lines.

Thereafter, additional local edge areas can be identified based on the possible parts of an identified edge area 10 which do not already form part of an identified local edge area 11. The procedure can also continue for all identified edge areas in the image, so that finally all identified edge areas have been divided into a number of local edge areas.

In a step 204, a second direction A2 is then identified for the local edge area 11. This second direction A2 can, in a manner similar to that of the first direction A1, be either related to the geometry of the image or the local edge area 11, for example to a longest diameter of the local edge area 11 or to an axis direction of the image.

In particular, the second direction A2 can be chosen to be the width in the first width direction discussed above or the same direction as the first direction A1. In the exemplary embodiment illustrated in Figure 4, the second direction A2 is selected to the width in the first width direction discussed above.

After this, in a step 205, 12 around the local edge area 11 is identified. an additional area The additional area 12 is illustrated in figure 4 with a dashed background. According to the invention, in identifying the additional area 12, pixels 14a, 14b which are located in immediate connection with a peripherally located boundary pixel 13 of the local edge area 11 and further away from the boundary pixel 13, along the second direction A2, up to a certain distance are selected. in the other direction A2 from the boundary pixel 13.

The term "boundary pixel" as used herein means a pixel which in the other direction A2 adjoins a pixel l4b which is not included in the local edge area 11. For a certain such edge pixel 13 the adjacent pixel l4b which is not included in the local edge area is thus added 11 to the additional area 12. In addition, the additional pixel 14a is added in the second direction A2, and so on, up to the said distance 15B 25B BBB 14 from the boundary pixel 13. According to the invention, the distance is proportional to the local edge area. ll width. in the second at the direction A2 between the boundary pixel 13. The ratio of the distance of the additional area 12 and the width of the local edge area 11 in the second direction A2 may be any suitable constant, and 5, according to a preferred embodiment between 1, preferably between 1 and 3, preferably 1. In the present exemplary embodiment, the constant 1 is used, which means that the distance is selected to be the same as the width in the other direction A2.

For the illustrated boundary pixel 13, the width at the boundary pixel 13 in the second direction A2 is two pixels, 14b, so that two further pixels 14a, the area 11, which is not included in the local edge, are selected to be included in the additional area 12 in the second direction. direction A2. But the size of the distance is thus variable over different boundary pixels. As is clear from Figure 4, the distance varies between 1 and 5 pixels for different boundary pixels.

According to a preferred embodiment, additional areas are identified in this way for all boundary pixels in the local edge area 11, and all additional areas thus identified together form the additional area for the local edge area 11.

This results in high repeatability in the identification of a characteristic part.

According to the invention, the characteristic part of the image is then identified so as to include the local edge area and the additional area. In other words, the characteristic part will constitute a coherent area of the image, comprising partly edge pixels and partly additional pixels. The characteristic portion is then stored in digital form on a storage medium, which may be any conventional storage medium as described above.

Thus, the method of the present invention is based on an edge area in an image.

Such edge areas can generally be automatically identified for different images in a highly repeatable manner. Since the interesting information in photographic images in many cases tends to be associated with areas of relatively high contrast and / or which form boundaries between fields present in the image, for example between foreground and background in the image, the edge areas identified by an automatic algorithm in general to be located in portions of the image that are interesting from, for example, a point of view of comparison. Particularly good repeatability has been achieved by the present inventors by means of the method described above for identifying an edge area. By selecting a local edge area with only a local maximum, spatially limited, distinct edge areas of suitable size can be obtained quickly and in a repeatable manner.

By identifying a characteristic part such as the union of an edge area and an additional area in the manner indicated above, the characteristic part will also include portions of the image surrounding an edge area. It has surprisingly been found that such surrounding areas often contain highly relevant image information. By selecting an additional area whose dimensions are proportional to the dimensions of the edge area, a characteristic part is obtained whose extent in an image becomes highly repeatable. In other words, approximately the same portion of the image with respect to the image information itself will be identified even if the image has been modified with respect to resolution, angular changes, focusing, rotation, color settings, etc. Since the edge area is identified only on the basis of the image content itself, and since the additional area is identified on the basis of the edge area, there are no dependencies on, for example, the external geometry of the image matrix.

Finally, this means that characteristic parts identified by a method according to the present invention are very likely to give repeatable results when used for a wide range of different purposes, such as automatic comparisons between images.

It is preferred that the digitally stored characteristic part be made to include information regarding which pixels constitute edge pixels, and also the original pixel information from the image. The storage of a characteristic part in this way means that the part can be used in a simple way in analyzes that also take into account which pixels constitute edge pixels. In many applications, this can be an interesting parameter in, for example, comparisons between images, since these pixels can, for example, mark boundaries between fields in the image.

According to a preferred embodiment, not only a first direction is selected, but two first directions. These directions may, in the same manner as described above for the first direction, be any directions related to the dimensions of the image or to the dimensions of the identified edge area, as long as the first two directions are not parallel. Then, the identification of the local edge area is performed in a corresponding manner as described above, but using both of the first directions separately. Thus, the local edge area is selected as a part of the identified edge area which only has a local maximum along both the respective first directions. This results in higher repeatability when identifying characteristic parts even in the case of the image being rotated and also in the case of elongated edge areas. It is preferred that the first two directions be perpendicular. Examples include the direction along the longest diameter of the edge area and its normal direction, respectively, as well as two perpendicular axis dimensions of the image. The latter in particular enables fast calculations in an automated procedure in combination with high consistency and repeatability.

In a similar manner, according to a preferred embodiment, not only a second direction but two other directions are selected. These two other directions can also, as described above for the other direction, be any directions that are related to the dimensions of the image or to the dimensions of the identified edge area. However, they must not be parallel. Thereafter, the identification of the additional area is performed in a corresponding manner as described above for the other direction, but using both the other directions.

Thus, an additional area is first identified by means of one of the other directions. Then, in a corresponding manner, an additional additional area is identified by means of the other of the other two directions. Finally, the additional area is selected for the characteristic part that the Union of these identified additional areas. Such a method provides further improved consistency and repeatability, since the additional area will not change its shape appreciably even in the case where the image has, for example, been rotated in relation to its original appearance.

If further improved repeatability is desired, it is further possible to use further first and second directions, respectively. According to a preferred embodiment 1, the image is a stand-alone photographic image or a photographic image included in a film sequence. According to another preferred embodiment, the image is a microscopic image or an image which is the result of a measurement of an object such as an internal structure, X-ray image of a patient or an article to be inspected for material defects, or alternatively a cross section of an object three-dimensional structure, such as an NMR image of tissue in a patient.

Preferred embodiments have been described above. However, it will be apparent to those skilled in the art that many changes may be made to the described embodiments without departing from the spirit of the invention. Thus, the invention should not be limited to the described embodiments, but may be varied within the scope of the appended claims.

Claims (12)

15 20 25 30 535 BBB 19 P A. T E N T K R .A. V Method for identifying a characteristic part of an image formed in the form of a matrix made up of pixels, drawn 8) b) C) (1), wherein the image (1) is stored in a digital sense - by the following steps : identifying an edge area (10) in the image (1) consisting of a continuous amount of edge pixels which together ((l0) in the image; identify a first direction (A1) in the image (1) the edge area (10): mans forms an edge area for identifying a local edge area (ll) consisting of a continuous set of pixels, for which locally (ll) along the first direction edge area the width in a first width direction (Al) maximum in addition to a single global maximum; (A2) lacks local identify a second direction for the local edge area (l1); identify an additional area (l1) (l2) around the local (l4a, 14b) located immediately adjacent to a peripheral (13) and further away from the boundary pixel edge area which includes pixels located as boundary pixel (11) of the local edge area (13) (A2) proportional to the local (II) at the boundary pixel (13): up to a distance in the other direction (13) of the edge area from the boundary pixel width in the other direction (A2) identify the characteristic part of the image (1 ) so as to include the local edge area (11) and (12), the part in digital form on a storage medium. additional area and store the characteristic 10 20 25 30 535 BBB 20 A method according to claim 1, characterized in that in step f), the digital storage of the characteristic part is made to include the original pixel information from the image and information regarding which pixels in the characteristic part constitute edge pixels. A method according to claim 1 or 2, (10) by a method comprising the steps of: characterized in that in step a) the edge area is made to be identified in. For all pairs of pixels in the image where one pixel (3a-3h) located in. a predetermined local environment (3) of the second pixel (4), determining the value of a comparison parameter which is the difference between the two pixels regarding an intensity property over one or more channels; ii. determine a threshold value (dl) (dl) minimum with respect to the comparison parameter as for the comparison parameter, where the threshold value is the first local mircomer in a function (f) which for each value of the comparison parameter indicates the number of compared pixel pairs in the image which in step i) showed this value for the comparison parameter, if the function (f) is first filtered to reduce the effect of noise in determining the first local minimum (d1); (10) area of edge pixels in which all edge pixels constitute iii. identify the edge area son) a coherence of pixels for which the comparison parameter when compared with another pixel in step i) was found to be greater than the threshold value (dl). A method according to claim 3, characterized in that steps i) and ii) are caused to be performed for each of a set of angularly evenly distributed, predetermined BBB 21 directions in the image, so that for each such predetermined direction first in step i) each pixel in the image is compared with an adjacent pixel along the direction, and then in step ii) a threshold value is determined for the pixel pairs compared in step i), and so that a respective threshold value is obtained for each predetermined direction , and by identifying the edge area (10) i as a coherent area Step iii) edge pixels where all edge pixels consist of pixels for which the comparison parameter when compared to another pixel in step i) was found to be greater than the threshold value of the other pixel direction for at least one of the predetermined directions. Method according to claim 4, characterized in that each edge pixel is associated with a binary represented number, in which each bit corresponds to a certain direction (0 or 1) depending on whether the comparison parameter of the edge pixel in question, and where the respective bit value relates in the direction corresponding to the bit was found to be greater than the threshold value for this direction, so that a binary number for all edge pixels in this way is caused to store information regarding along which directions the edge pixel in question meets the criterion of being an edge pixel. Method according to any one of claims 3-5, characterized in that the intensity property of a pixel is the total luminance of the pixel in question. A method according to claim 6, that the image (1) is characterized in a first step preceding step a) is caused to be converted to a grayscale image, in which the total luminance of a pixel is constituted by the value of the grayscale of the pixel in question. 10 15 20 25 30 535 888 22 Method according to one of the preceding claims, characterized in that step e) is caused to be performed for all (11) borders on a pixel in the image (1) (10), the part of the image (1) is made to include all identifications. boundary pixels in the local edge area (A2) sonx in the other direction S0m are not included in the edge area and that the characteristic added areas to the local edge area (ll). A method according to any one of the preceding claims, characterized in that in step b) two different first directions are selected, and in that in step c) the local edge area (ll) is selected so that it lacks local maxima in addition to a respective global maximum in both the first directions separately. 10. A method according to claim 9, characterized in that the first two directions are made to consist of two mutually perpendicular coordinate axes in the image matrix. 11. ll. Procedure according to. any of the preceding claims, characterized in that in step d) two different other directions are caused to be selected, (12) and in that in step e) the additional area is caused to be identified comprising pixels which are located along each respective other direction in immediate connection to a boundary pixel of the local edge area (11) and further away from the boundary pixel up to a distance in the respective other direction from the boundary pixel which is proportional to the width of the local edge area (II) in the respective other direction at the boundary pixel. 12. A method according to claim 11, characterized in that the two other directions are made to consist of two perpendicular coordinate axes in the image matrix.