WO2015010275A1

WO2015010275A1 - Interest point judgement method and interest point judgement device

Info

Publication number: WO2015010275A1
Application number: PCT/CN2013/080007
Authority: WO
Inventors: 周强; 刘峥; 许国军
Original assignee: 华为技术有限公司
Priority date: 2013-07-24
Filing date: 2013-07-24
Publication date: 2015-01-29
Also published as: CN104541289A; CN104541289B

Abstract

Provided is an interest point judgement method, which is used for judging whether a current pixel is an interest point of a current image or not. The method comprises: comparing the amplitude of a current pixel with the amplitude of other pixels in a local area where the current pixel is located, and when the amplitude of the current pixel is an extremum value, determining an area on a target image corresponding to the local area as a target area;using a second filtration parameter group to filter the target area, so as to obtain a second area;comparing the amplitude of the current pixel with the amplitude of all the pixels in the second area, and when the amplitude of the current pixel is an extremum value, using a third filtration parameter group to filter the target area, so as to obtain a third area;and comparing the amplitude of the current pixel with the amplitude of all the pixels in the third area, and when the amplitude of the current pixel is an extremum value, determining the current pixel as the interest point of the current image.Also provided is an interest point judgement device.The embodiments of the present invention can greatly reduce the occupation of a memory in an interest point judgement process.

Description

Interest point judging method and interest point judging device

The present invention relates to a point of interest detection technique, and more particularly to a point of interest determination method and a point of interest determination apparatus. Background technique

Image feature extraction is one of the key technologies in the field of image recognition. The core of this technology lies in the detection of Interest Point.

1 is an exemplary flow chart of an existing point of interest detection method 100.

Step 102: Establish a LoG (Laplacian of Gaussian) filtered image pyramid (Image Pyramid) of the original image.

2 is an exemplary flow chart of a prior art LoG filtered image pyramid generation method 200.

Step 202: Perform continuous down sampling (Down Sample) on the original image to obtain multiple target images.

Specifically, first, the original image is downsampled to generate a target image 1. Next, the target image 1 is downsampled to generate a target image 2. Again, the target image 2 is downsampled to generate a target image 3. The above process continues until the target image M is generated. Among them, the original image can be regarded as the target image 0.

In the specific implementation process, the above-mentioned downsampling refers to reducing the image to be targeted according to the aspect ratio. Specifically, a common image down sampling method such as a neighbor sampling method or a bilinear interpolation method can be used. After downsampling, the size of the target image M may be, for example but not limited to, 1/2 of the target image M-1.

Step 204: Perform multiple LoG filtering on each target image to generate multiple LoG filtered images of the target image.

Specifically, in the process of performing LoG filtering on the target image, the target image is first performed. Gaussian filtering to generate a Gaussian filtered image. Next, Laplace filtering is performed on the Gaussian filtered image to generate a LoG filtered image.

The Gaussian filter parameters used by Gaussian filtering can be expressed as σ(Ν) during each LoG filtering process. For the same target image, in the three L 0 G filtered images obtained by using σ(Ν - 1), σ(Ν) and σ(Ν + 1) as Gaussian filter parameters, the Gaussian filter parameter is σ( Ν - 1) The L 0 G filtered image is the upper layer image of the LoG filtered image with Gaussian filter parameter _σ (Ν), and the LoG filtered image with Gaussian filter parameter σ(Ν + 1) is the LoG with Gaussian filter parameter σ(Ν) Filter the underlying image of the image. That is to say, a LoG filtered image and its upper layer image and lower layer image are both derived from the same target image, and the Gaussian filtering parameters used to generate the LoG filtered image and its upper layer image and the lower layer image are σ(Ν), σ( Ν - 1) and σ(Ν + 1). Furthermore, the Laplacian filter parameters used to generate different LoG filtered images may be different. It can be seen that multiple LoG filtered images of the same target image are sequentially generated in a certain order, and the order can be expressed by Gaussian filtering parameters, that is, multiple LoG filtered images sequentially generated, which are used in the generation process. The Gaussian filtering parameters are σ(1), σ(2), σ(3) ... σ(Ν - 1), σ(Ν) in order, and these LoG filtered images are continuous with each other. In the multiple LoG filtered images generated for a target image, the LoG filtered image generated based on the Gaussian filter parameter σ(Ν) is the Nth layer LoG filtered image in these LoG filtered images.

After the above steps 202 and 204, a LoG filtered image pyramid can be generated. It is not difficult to find that the LoG filtered image pyramid includes a continuous plurality of sets of images, each set of images includes a plurality of consecutive LoG filtered images, and each set of images is a downsampled image of the previous set of images. In the specific implementation process, the number of LoG filtered images in each group of images can be set according to specific needs. Normally, each set of images contains at least three LoG filtered images. It should be noted that in the specific implementation process, the above steps 202 and 204 may also be performed by crossover, that is, each time a target image is generated, multiple Lo G filtering is performed on the target image to generate multiple L 0 G filters of the target image. image.

In addition, in a specific implementation process, another method described below may also be used to generate a LoG filtered image pyramid. First, the original image is subjected to multiple LoG filtering, which is the original image. The image is generated as a set of Lo G filtered images (the set of Lo G filtered images contains multiple Lo G filtered images of the original image). Thereafter, each LoG filtered image in the set of LoG filtered images of the original image is separately downsampled to generate another set of LoG filtered images. Thereafter, each LoG filtered image in the other set of LoG filtered images is downsampled to generate a further set of LoG filtered images. The above process continues until M sets of LoG filtered images are generated. The M sets of LoG filtered images generated in the above sequence can form a LoG filtered image pyramid.

FIG. 3 is an exemplary schematic diagram of a conventional LoG filtered image pyramid generation process. As shown in FIG. 3, the original image is subjected to downsampling to generate a target image 1, and the target image 1 is subjected to downsampling to generate a target image 2, wherein the target image 1 is 1/2 of the original image, and the target image 2 is 1/ of the target image 1. 2.

Each target image (including the original image) is subjected to three LoG filtering to generate three LoG filtered images of the target image. Wherein, each LoG filtering process includes first performing Gaussian filtering on the target image to generate a Gaussian filtered image. Then, the Gaussian filtered image is subjected to Laplacian filtering to generate a LoG filtered image. Taking the target image 1 as an example, after three LoG filtering, three LoG filtered images 302-306 are finally generated. In the three LoG filtered images 302-306 obtained for the same target image (ie, the target image 1), the Gaussian filter parameter used to generate the LoG filtered image 302 is σ(1), which is used to generate the LoG filtered image 304. The Gaussian filter parameter is σ(2), and the Gaussian filter parameter used to generate the LoG filtered image 306 is σ(3). Therefore, the LoG filtered image 302 is the upper layer image of the LoG filtered image 304, and the LoG filtered image 306 is the LoG filtered image 304. The underlying image.

After the above processing, three sets of images from bottom to top will be generated. The first set of images includes three LoG filtered images generated by performing three LoG filtering on the original image, and the second set of images includes three times on the target image 1. Three LoG filtered images generated by LoG filtering, and the third set of images includes three LoG filtered images generated by performing three LoG filtering on the target image 2. The three sets of images constitute a LoG filtered image pyramid 308.

In the specific implementation process, the number of times of downsampling and the number of LoG filtered images can be rooted. Set according to specific needs.

The other steps in method 100 continue below.

After performing step 102 in method 100, at step 104, points of interest for each LoG filtered image are determined.

FIG. 4 is an exemplary schematic diagram of an existing point of interest judging process. As shown in Fig. 4, three LoG filtered images are shown, which are obtained by performing three LoG filtering on the same target image. For ease of description, the three LoG filtered images are the three LoG filtered images 302-306 in Figure 3. As described above, the LoG filtered images 302-306 are ultimately generated by performing three LoG filtering on the target image 1 in Fig. 3. In the three LoG filtered images 302-306, the Gaussian filter parameter used to generate the LoG filtered image 302 is σ(1), and the Gaussian filter parameter used to generate the LoG filtered image 304 is σ(2), and a LoG filtered image is generated. The Gaussian filter parameter used by 306 is σ(3). Therefore, the LoG filtered image 302 is the upper layer image of the LoG filtered image 304, and the LoG filtered image 306 is the lower layer image of the LoG filtered image 304. The following takes pixel 314 as an example to introduce the judgment process of interest points of LoG filtered images.

Specifically, when it is determined whether the pixel 314 is a point of interest of the LoG filtered image 304, the amplitude of the pixel 314 needs to be compared with the amplitude of at least 26 other pixels. If the comparison result shows that the amplitude of the pixel 314 is an extreme value (Extremum, such as a maximum value or a minimum value), the determination pixel 314 is a point of interest of the LoG filtered image 304. To describe the specific locations of the 26 other pixels, a local area is first defined on the LoG filtered image 304, the local area including at least pixels 314 and 8 pixels around the pixel 314, such as the local area 310 on the LoG filtered image 304. The eight pixels around the pixel 314 are respectively pixels 316-330. That is, the local area 310 is a 3x3 area centered on the pixel 314. In a specific implementation process, the local area 310 may also adopt an NxN area centered on the pixel 314, where N is an odd number greater than 3. Next, the corresponding area on the upper layer image (i.e., LoG filtered image 302) and the lower layer image (i.e., LoG filtered image 306) of the LoG filtered image 304, that is, the upper layer area 308 and the lower layer area 312 are determined. Specifically, the location of the local area 310 on the LoG filtered image 304 is first determined. The coordinates, and then the areas indicated by the above position coordinates on the LoG filtered image 302 (upper layer image of the LoG filtered image 304) and 306 (lower layer image of the LoG filtered image 304), that is, the upper layer area 308 and the lower layer area 312 are determined. As shown in FIG. 4, the position of the upper layer region 308 on the LoG filtered image 302 (the upper layer image of the LoG filtered image 304) is the same as the position of the local region 310 on the LoG filtered image 304, and the lower layer region 312 is at the LoG filtered image 306 (LoG). The position on the lower layer image of the filtered image 304 is the same as the position of the local area 310 on the LoG filtered image 304. In this way, the 26 other pixels are 8 pixels out of the pixel 314 in the local area 310, 9 pixels 332-348 in the upper layer area 308, and 9 pixels 350-366 in the lower layer area 312. As can be seen from the above process, when determining whether a pixel is a point of interest of a LoG filtered image in which the pixel is located, it is necessary to simultaneously use an upper layer image and a lower layer image of the LoG filtered image in which the pixel is located. As described above, the LoG filtered image and the upper layer image and the lower layer image are both derived from the same target image, and the Gaussian filtering parameters used when generating the LoG filtered image and the upper layer image and the lower layer image are respectively σ(Ν) , σ(Ν _ΐ;^.σ(Ν + 1).

It is not difficult to understand that the upper and lower images of the LoG filtered image are used in determining the point of interest of each LoG filtered image. Since the first L 0 G filtered image in each set of LoG filtered images in the LoG filtered image pyramid has no upper layer image, the last L 0 G filtered image has no lower layer image. Therefore, in the specific implementation process, it is often only determined. The interest points of the first LoG filtered image and the other LoG filtered images other than the last LoG filtered image in each group of LoG filtered images may be used. Of course, some existing methods can also be used to determine the points of interest of the first LoG filtered image and the last LoG filtered image. In addition, determining whether a pixel is a point of interest of a LoG filtered image in which a pixel is located requires at least 8 pixels around the pixel. When the pixel to be judged is at the edge of the image it is in, there may be fewer than eight pixels around the pixel. In this case, the points of interest at the edges of the image may not be judged, and of course, some methods may be used to determine whether the pixels are points of interest of the image in which they are located.

The other steps in method 100 continue below.

After performing step 104 in method 100, at step 106, according to step 104 Determine the points of interest of each LoG filtered image and determine the points of interest of the original image.

Specifically, various methods can be used to determine the points of interest of the original image based on the interest points of each LoG filtered image.

It should be noted that, in the specific implementation process, if the first LoG filtered image and the last LoG filtered image of each set of LoG filtered images in the LoG filtered image pyramid are not determined by their respective points of interest, the original image is determined. In the process of interest points, these LoG filtered images without determining the points of interest are not considered.

It is not difficult to be found by those skilled in the art that in determining the point of interest of a LoG filtered image, in the process of determining the point of interest shown in FIG. 4, three LoG images need to be loaded in the memory simultaneously, that is, the LoG to be determined. The filtered image and the upper layer image and the lower layer image of the LoG filtered image. This makes the mobile terminal have some difficulty in performing the above operations. It is well known that the performance of cameras of mobile terminals (such as various smart phones) is getting stronger and stronger, and the images taken are becoming more and more clear, which leads to an increasing space for each picture. In this case, in order to determine a point of interest of a LoG filtered image (for example, in the process of using a mobile terminal for image comparison), it is necessary to simultaneously load three images in the memory of the mobile terminal, which inevitably takes up A large amount of valuable memory resources affect the overall performance of the mobile terminal. Summary of the invention

In view of this, it is necessary to provide a method of judging the point of interest to solve the problem that the existing point of interest judging method takes up too much memory.

At the same time, a point of interest judging device is provided to solve the problem that the existing point of interest judging method takes up too much memory.

According to an aspect of the present invention, a method for determining a point of interest is provided for determining whether a current pixel is a point of interest of a current image in which a current pixel is located, wherein the current image is performed on a target image by using a first filter parameter set Obtained by filtering, the method includes:

Comparing the amplitude of the current pixel with the amplitude of other pixels in the local region where the current pixel is located on the current image, and determining, when the comparison result shows that the amplitude of the current pixel is an extreme value, a corresponding area of the local area on the target image as a target area;

And filtering the target area by using a second filter parameter set to obtain a second area; comparing an amplitude of the current pixel with an amplitude of all pixels in the second area, and comparing the result, the amplitude of the current pixel is one pole And a third filter parameter group is used to filter the target area to obtain a third area;

The amplitude of the current pixel is compared with the amplitude of all the pixels in the third region, and when the comparison result shows that the amplitude of the current pixel is an extreme value, it is determined that the current pixel is the point of interest of the current image.

According to another aspect of the present invention, a point of interest judging device is provided for determining whether a current pixel is a point of interest of a current image in which a current pixel is located, wherein the current image is a target image by using a first filter parameter group The filtering process is performed, and the device includes:

a memory, configured to store the current image;

The processor is configured to perform the following operations:

Comparing the amplitude of the current pixel with the amplitude of other pixels in the local area where the current pixel is located on the current image, and determining that the local area is on the target image when the comparison result shows that the amplitude of the current pixel is an extreme value a corresponding area, as a target area; filtering the target area using a second filter parameter set to obtain a second area;

Comparing the amplitude of the current pixel with the amplitude of all the pixels in the second region, and when the comparison result shows that the amplitude of the current pixel is an extreme value, filtering the target region by using the third filter parameter group to obtain the first Three regions;

When determining whether a pixel is a point of interest of an image of the pixel, the embodiment of the present invention does not need to simultaneously load the entire upper layer image and the entire lower layer image of the image in the memory, but only needs to temporarily calculate the local area where the pixel is located. Corresponding regions on the upper layer image and the lower layer image. It can be seen that the technical solution provided by the embodiment of the present invention can greatly reduce the occupation of memory by the point of interest judging process. DRAWINGS

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings to be used in the embodiments of the present invention will be briefly described below. It is obvious that the drawings described below are only some embodiments of the present invention, Those skilled in the art can also obtain other drawings based on these drawings without paying any creative work.

1 is an exemplary flowchart of a method for detecting a point of interest;

2 is an exemplary flowchart of a conventional Lo G filtered image pyramid generation method;

3 is an exemplary schematic diagram of a conventional LoG filtered image pyramid generation process;

4 is an exemplary schematic diagram of a prior interest point determination process;

FIG. 5 is an exemplary flowchart of a method for judging a point of interest according to an embodiment of the present invention; FIG. 6 is a schematic diagram of a process of performing a LoG filtering operation on a target area according to an embodiment of the invention;

7 is a schematic diagram of a process of a reverse symmetric filling method;

8 is an exemplary schematic diagram of a point of interest judging process according to an embodiment of the present invention; FIG. 9 is an exemplary flowchart of a method for judging a point of interest according to an embodiment of the present invention; FIG. 10 is an illustration of an interest according to an embodiment of the present invention. FIG. 11 is a schematic diagram showing an exemplary hardware structure of a point of interest judging apparatus according to an embodiment of the present invention; FIG.

FIG. 12 is a schematic diagram showing an exemplary hardware configuration of a point of interest judging apparatus according to an embodiment of the present invention; FIG.

FIG. 13 is a schematic diagram showing an exemplary logical structure of a point of interest judging apparatus according to an embodiment of the present invention; FIG.

Figure 14 is a diagram showing an exemplary logical structure of a point of interest judging device in accordance with an embodiment of the present invention. detailed description

FIG. 5 is an exemplary flow chart of a point of interest determination method 500 in accordance with an embodiment of the present invention. Square The method 500 is configured to determine whether the current pixel is a point of interest of a current image where the current pixel is located, wherein the current image is obtained by filtering the target image by using the first filter parameter group.

Step 502: Comparing the amplitude of the current pixel with the amplitude of other pixels in the local area where the current pixel is located on the current image, and determining, when the comparison result shows that the amplitude of the current pixel is an extreme value, determining the local area in the The corresponding area on the target image serves as the target area.

The contents of the technical features such as the local area and the extreme value have been described in detail in the background art, and therefore will not be described herein. In addition, in the specific implementation process, when comparing the current pixel with other pixels in the local area, other pixel attributes other than the amplitude may be compared, such as but not limited to gray values. In summary, when comparing pixels, the content of the comparison can be collectively referred to as the pixel value.

In determining the target area, the positional coordinates of the local area on the current image may be first determined, such as determining the positional coordinates of the boundary of the local area on the current image. Then, an area indicated by the position coordinates on the target image, that is, an area surrounded by the boundary coordinates on the target image is determined, and a corresponding area of the partial area on the target image can be obtained.

Step 504: Perform filtering processing on the target area by using a second filter parameter group to obtain a second area.

Step 506, comparing the amplitude of the current pixel with the amplitude of all the pixels in the second region, and when the comparison result shows that the amplitude of the current pixel is an extreme value, filtering the target region by using the third filtering parameter group. , get the third area.

In particular, each filter parameter set (e.g., the first filter parameter set, the second filter parameter set, and the third filter parameter set) may include one or more filter parameters. In the following description, the description is made by taking two filtering parameters for each filtering parameter group as an example.

When each filter parameter set includes two filter parameters, the two filter parameters may be respectively recorded as a first filter parameter and a second filter parameter. In this case, for an image to be processed, the filtering process may include: performing a first filtering operation on the image to be processed using the first filtering parameter in the filtering parameter group used in the filtering process to obtain a first filtered image; The filtering process used The second filtering parameter in the filtering parameter group performs a second filtering operation on the first filtered image to obtain a second filtered image. Since the filtering processing according to the first filtering parameter group, the second filtering parameter group, and the third filtering parameter group is performed for the same target image, if the first filtering parameter in the first filtering parameter group is set For _σ ( N ), the first filter parameter in the second filter parameter set is set to σ(Ν - 1), and the first filter parameter in the third filter parameter set is set to σ(Ν + 1 The second area can be regarded as the upper layer area of the partial area (that is, the second area can be regarded as the corresponding area of the upper area image of the current image in the current image), and the third area can be regarded as the lower layer area of the partial area ( That is, the third area can be regarded as the corresponding area of the above-mentioned partial area in the lower layer image of the current image. For another example, since the filtering process according to the first filter parameter group, the second filter parameter group, and the third filter parameter group is performed for the same target image, if the first filter parameter group is the first one The filter parameter is set to σ(Ν), the first filter parameter in the second filter parameter set is set to σ( Ν + 1), and the first filter parameter in the third filter parameter set is set to σ ( Ν - 1), the second area can be regarded as the lower layer area of the local area (ie, the second area can be regarded as the corresponding area of the local image in the lower layer image of the current image), and the third area can be regarded as the partial area The upper region (ie, the third region can be regarded as the corresponding region of the upper region image of the current image in the current image). However, it should be noted that, in a specific implementation process, the filtered image obtained by separately filtering the same target image using the first filter parameter group, the second filter parameter group, and the third filter parameter group may also be mutually Continuously, as long as the first filter parameter in the first filter parameter group, the second filter parameter group, and the third filter parameter group are different. For example, the first filter parameter in the first filter parameter group is σ(Ν), the first filter parameter in the second filter parameter group is σ(Ν -3), and the first filter parameter in the third filter parameter group is σ(Ν + 5). Or the first filter parameter in the first filter parameter group is σ(Ν), the first filter parameter in the second filter parameter group is σ(Ν + 2), and the first filter parameter in the third filter parameter group is σ(Ν - 4). Or the first filter parameter in the first filter parameter group is σ(Ν), the first filter parameter in the second filter parameter group is σ(Ν - 2), and the first filter parameter in the third filter parameter group is σ(Ν -5). Or the first filter parameter in the first filter parameter group is σ(Ν), and the first filter parameter in the second filter parameter group is σ(Ν + 2), the first filter parameter in the third filter parameter set is σ(Ν + 5). Of course, if the filtered images obtained by separately filtering the same target image using the first filter parameter set, the second filter parameter set, and the third filter parameter set are continuous with each other, the determined interest points are more accurate. .

In a specific implementation process, the first filtering parameter may be a Gaussian filtering parameter, the second filtering parameter may be a Laplacian filtering parameter, the first filtering operation is a Gaussian filtering operation, and the second filtering operation is a Laplacing operation. Filter operation. In this way, the first filtered image is a Gaussian filtered image, and the second filtered image is a LoG filtered image. Certainly, the foregoing first filtering parameter may also be a filtering parameter used by other filtering operations than the Gaussian filtering operation, and the second filtering parameter may also be a filtering parameter used by other filtering operations than the Laplacian filtering operation. . The first filtering operation may be other filtering operations than the Gaussian filtering operation, and the second filtering operation may be other filtering operations than the Laplacian filtering operation.

In the specific implementation process, the following setting may be further made, that is, _CT (N) = k ^N j, where k and j are constants. Therefore, the first filtering parameter in the first filtering parameter group is (N) = k ^N j , and the first filtering parameter in the second filtering parameter group is ^-1 j or k ^N+1 j, and the third filtering The first filter parameter in the parameter group is k ^N+1 j or! ^. In the specific implementation process, the values of k and j can be set according to experience and specific needs, such as k = , j = 1.6. When the first filtering parameter is a Gaussian filtering parameter, σ(Ν) is generally referred to as a Gaussian filtering kernel. The Gaussian filtering process convolves the Gaussian function with the image to be filtered, ie G(x, y) = g(x, y) * I(x, y) , where g(x, y) is a Gaussian function, g (x, y) = ^_e" ² 3⁄4 , I(x, y) is the image matrix of the image to be filtered. Meanwhile, based on the above settings, when the second filter parameter is a Laplacian filter parameter, the scale normalization pull Plass filtering operation

0 1 0 The Laplacian filter template used in the process can be, for example but not limited to, 1 - 4 1 or

0 1 0

1 1 1

σ 1 - 8 1 and so on, at this time, σ ^{2 is} called a scale normalization factor. In this case, the scale normalization factor of the Laplacian filter template used to generate the current image 1 1 1 image is (k ^N j) ² , and the second region is generated. The scale normalization factor of the Laplacian filter template used is (k^j) ² or (k ^N+1 j) ² , and the scale normalization factor of the Laplacian filter template used to generate the third region is (k ^N "j) ² or (k ^N - ¹ j ) ^2. At this time, it can be understood that the Gaussian filter parameter and the Laplacian filter parameter are both σ(Ν). However, those skilled in the art It should be understood that in a specific implementation process, the Gaussian filter parameters and the Laplacian filter parameters in the same filter parameter group may be separately set, and there may be no association between the two.

FIG. 6 is a schematic diagram of a process of performing a LoG filtering operation on a target area according to an embodiment of the invention. In the scenario shown in FIG. 6, the target image 600 includes a target area 602, the central pixel of the target area 602 is the pixel 604, the size of the area is 3x3, and the filtering window size of the Gaussian filter is 5x5, Laplacian filtering The filter window size is 3x3.

Gaussian filtering of the target region 602 in accordance with the Gaussian filtering principle requires the use of a region 606 of the target image 600 centered on the pixel 604 and having a size of 7x7.

After performing Gaussian filtering on the target region 602, if Laplace filtering is performed on the Gaussian filtered target region 602, a region having a size of 5x5 is required, and the region is passed through the pixel 604 on the target image 600. It is obtained by Gaussian filtering of the center 608 of the size 5x5. According to the Gaussian filtering principle, Gaussian filtering of the region 608 requires the use of a region 610 of the target image 600 centered on the pixel 604 and having a size of 9x9.

It can be seen that to obtain the LoG filtering region of the target region 602, a 9x9 sized region 610 centered on the pixel 604 on the target image 600 is needed, and the region 610 is larger than the target region 602. In the specific implementation process, the filter window size of the Gaussian filter is ΝχΝ, where Ν is an odd number greater than or equal to 3, and the specific value can be set according to specific needs. According to the principle of Laplacian filtering, the filtering window size of Laplacian filtering is usually 3x3. It can be seen that the size of the area 610 is basically determined by the filtering window size of the Gaussian filtering.

If a portion of region 610 is beyond the boundary of target image 600, the pixels may be filled using, for example, but not limited to, a reverse symmetric fill method. The reverse symmetric filling method will be briefly described below with reference to FIG. Figure 7 is a schematic diagram of the process of the reverse symmetric filling method. The boundary portion of the target image in which the target area is located is as shown in area 702 in FIG. As shown in FIG. 7, the upper boundary of the region 702 is the boundary 706, the lower boundary is the boundary 708, and the right boundary is the boundary 704. If a padding region 702 containing two columns of pixels (ie, column 列 and column 2) is filled on the right side of the region 702 using the inverse symmetric filling method, the value of the column 可以 can take the value of column 1, column 2, The value of column 2 can be taken as the value of column 2, that is, the two columns of pixels (ie, column 1, and column 2) in padding region 702, and the two columns of pixels in region 702 (ie, column 1 and column 2) are bounded by boundary 704. bilateral symmetry. Alternatively, the value of the column can take the value of column 2, and the value of column 2 can take the value of column 3, that is, the two columns of pixels (ie, column 1, and column 2) in padding region 702, and region 702 The two columns of pixels (ie, column 2 and column 3) are bilaterally symmetric with column 1 as the axis.

The other steps in method 500 continue below.

After performing step 506 in method 500, in step 508, comparing the amplitude of the current pixel with the amplitude of all pixels in the third region, and determining the current when the comparison result shows that the amplitude of the current pixel is an extreme value The pixel is the point of interest of the current image.

In a specific implementation process, the extreme values described in steps 502, 506, and 508 may be the same as the maximum value, or the same as the minimum value. Specifically, in step 502, if the amplitude of the current pixel is a maximum value when compared with the amplitudes of other pixels in the local area, the criterion in step 506 is that the amplitude of the current pixel is The amplitudes of all the pixels in the second region are also a maximum value when compared. The criterion in step 508 is that the amplitude of the current pixel is still a large value when compared with the amplitude of all the pixels in the third region. value. Or, if the amplitude of the current pixel is a minimum value when compared with the amplitudes of other pixels in the local area, the criterion in the step 506 is that the amplitude of the current pixel is in the second region. The amplitude of the pixel is also a small value when compared. The criterion in step 508 is that the amplitude of the current pixel is still a small value when compared with the amplitude of all pixels in the third region.

When determining whether a pixel is a point of interest of an image of the pixel, the embodiment of the present invention does not need to simultaneously load the entire upper layer image and the entire lower layer image of the image in the memory, but only needs to temporarily calculate the local area where the pixel is located. Corresponding regions on the upper layer image and the lower layer image. It can be seen that the technical solution provided by the embodiment of the present invention can greatly reduce the memory occupied by the point of interest judging process. use.

With the technical solution provided by the embodiment of the present invention, the user can use the smart terminal to perform image recognition, thereby performing operations such as price comparison of commodities. For example, when the user wants to compare the price of a certain item in another mall or online store in the mall, the user can take a photo of the item, and then use the smart terminal to extract the feature data of the photo and transmit it to the background server via the Internet, the background server according to photos of feature data, stored in the database to match a large number of features of the product image feature data, the query to match the commodity, the price of goods re ^ ¹ match returned to the user.

FIG. 8 is an exemplary schematic diagram of a point of interest determination process in accordance with an embodiment of the present invention. As shown in Fig. 8, there is shown a LoG filtered image 802 obtained by LoG filtering a target image.

When determining whether the pixel 810 on the LoG filtered image 802 is a point of interest of the LoG filtered image 802, according to the technical solution provided by the embodiment of the present invention, first, the amplitude of the pixel 810 and the local area where the pixel 810 is located on the LoG filtered image 802. (In this embodiment, the local area is a 3x3 area) 804 compares the amplitudes of other pixels in the 804. When the amplitude of the comparison result display pixel 810 is an extreme value, determining the local area on the target image Corresponding area, as the target area.

Subsequently, the target region is subjected to filtering processing using a LoG filter parameter set constructing a layer image of the LoG filtered image 802 to obtain an upper region 806 of the local region 804.

Subsequently, the amplitude of the pixel 810 is compared with the amplitude of all pixels in the upper layer region 806. When the amplitude of the comparison result display pixel 810 is an extreme value, the LoG filter parameter group constructing the lower layer image of the LoG filtered image 802 is used for the target. The region is subjected to filtering processing to obtain a lower region 808 of the local region 804.

Finally, the amplitude of the pixel 810 is compared with the amplitude of all pixels in the lower layer region 808. When the comparison results show that the amplitude of the pixel 810 is an extreme value, the pixel 810 is determined to be the point of interest of the LoG filtered image 802.

As can be seen from FIG. 8, when it is determined whether the pixel 810 is a point of interest of the LoG filtered image 802, it is only necessary to load the LoG filtered image 802 in the memory, and it is not necessary to simultaneously load the LoG filtered image 802. The entire upper layer image and the entire lower layer image are only required to temporarily calculate the corresponding regions 806 and 808 of the partial region 804 where the pixel 810 is located on the upper layer image and the lower layer image. It can be seen that the technical solution provided by the embodiment of the present invention can greatly reduce the occupation of memory by the point of interest judging process.

In the point of interest judging process shown in Fig. 8, the upper layer area 806 of the partial area 804 is first generated, and the lower layer area 808 of the partial area 804 is generated. However, in a specific implementation process, the lower region 808 of the local region 804 may also be generated first, and the upper region 806 of the local region 804 may be generated. Further, as a suboptimal point of interest judging scheme, the upper layer area 806 and the lower layer area 808 of the partial area 804 can also be generated at the same time. In this case, it is necessary to simultaneously compare the amplitude of the pixel 810 with the amplitudes of all the pixels in the upper layer region 806 and the lower layer region 808. It is not difficult to understand that this suboptimal point of interest judgment scheme will add additional calculations in some cases. For example, according to the point of interest judging process shown in FIG. 8, if the amplitude of the pixel 810 is no longer an extreme value when compared with the amplitude of all pixels in the upper layer area 806, the point of interest determination process according to FIG. It is no longer necessary to compare the amplitude of the pixel 810 with the amplitude of all pixels in the lower layer region 808, so there is no need to generate the lower layer region 808. According to the above suboptimal point of interest judging scheme, the lower layer area 808 still needs to be generated. It can be seen that compared with the point of interest judging method shown in FIG. 5 and FIG. 8, the above suboptimal point of interest judging scheme adds extra calculation amount in some cases.

It should be noted that in the specific implementation process, the method shown in Figure 5 is not only optimized for the LoG filtering scene. In the specific implementation process, the SURF (Speeded Up Robust Features) algorithm can be optimized based on the principle of the method shown in FIG. The optimized SURF algorithm is described below.

FIG. 9 is an exemplary flow diagram of a point of interest determination method 900 in accordance with an embodiment of the present invention. The method 900 is configured to determine whether the current pixel is a point of interest of a current image where the current pixel is located, where the current image is filtered by using the first block filter parameter, and then calculating the sea by the filtered target image. According to the Hessian determinant response, the current pixel is a pixel whose positive response value is on the current image.

Step 902: Comparing the amplitude of the current pixel with the amplitude of other pixels in the local area where the current pixel is located on the current image, and displaying the amplitude of the current pixel as a maximum value in the comparison result. And determining a corresponding area of the local area on the target image as a target area; Step 904, filtering the target area by using a second block filter parameter, and calculating Hessian on the filtered target area. Deterministic response, obtaining a second region;

Step 906, comparing the amplitude of the current pixel with the amplitude of all the pixels in the second region, and filtering the target region by using a third-party frame filtering parameter when the comparison result shows that the amplitude of the current pixel is a maximum value. Processing, and calculating a Hessian determinant response to the filtered target region to obtain a third region;

Step 908: Comparing the amplitude of the current pixel with the amplitude of all the pixels in the third region, and determining that the current pixel is the interest point of the current image when the comparison result shows that the amplitude of the current pixel is a maximum value.

In a specific implementation process, the first block filter parameter, the second block filter parameter, and the third-party frame filter parameter may correspond to different size filter matrices, and the second region is an upper region or a lower region of the local region, and the third region It is the lower area or the upper area of the target area.

FIG. 10 is an exemplary flowchart of a point of interest determination method 1000 in accordance with an embodiment of the present invention. The method 1000 is used to determine whether the current pixel is a point of interest of a current image where the current pixel is located, where the current image is obtained by filtering the target image by using the first filter parameter group, and is described on the current image. The amplitude of the current pixel is an extreme value compared to the amplitude of other pixels in the local area where the current pixel is located.

Step 1002: Determine a corresponding area of the local area on the target image as a target area.

Step 1004: Perform filtering processing on the target area by using a second filter parameter group to obtain a second area.

Step 1006: Comparing the amplitude of the current pixel with the amplitude of all the pixels in the second region, and filtering the target region by using the third filter parameter group when the comparison result shows that the amplitude of the current pixel is an extreme value. , get the third area.

Step 1008: Comparing the amplitude of the current pixel with the amplitude of all the pixels in the third region, and determining that the current pixel is the current image when the comparison result shows that the amplitude of the current pixel is an extreme value. Points of interest.

As described above, each filter parameter set includes a first filter parameter and a second filter parameter, and the filter process includes:

Performing a first filtering operation on the image to be processed using the first filtering parameter in the filtering parameter set used by the filtering process to obtain a first filtered image;

The second filtering operation is performed on the first filtered image using the second filtering parameter in the filtering parameter set used by the filtering process to obtain a second filtered image.

As described above, the first filtering parameter is a Gaussian filtering parameter, the second filtering parameter is a Laplacian filtering parameter, the first filtering operation is a Gaussian filtering operation, and the second filtering operation is a Lapu filter. Lass filtering operation.

As described above, the first filter parameter in the first filter parameter set is σ(Ν), and the first filter parameter in the second filter parameter set is σ( Ν + 1), the third filter The first filter parameter in the parameter set is σ(Ν -1) ; or

The first filter parameter in the first filter parameter set is _σ (Ν), and the first filter parameter in the second filter parameter set is σ(Ν -1), in the third filter parameter group The first filter parameter is σ(Ν + 1).

As described above, the following settings can be made, namely _CT (N) = k ^N _CT , where k and j are constants. As described above, the partial area includes at least a current pixel and 8 pixels adjacent to the current pixel.

The details of each technical feature have been described in detail in the foregoing, and therefore will not be described again here. In the specific implementation process, the pixel with the amplitude of one extreme value compared with the amplitude of other pixels in the local area may be firstly filtered on the current image, and then the method shown in FIG. 10 is applied to each pixel selected. 1000.

In a specific implementation process, the prior art can be combined with the method 1000 shown in FIG. 10 to determine a point of interest of the current image. Specifically, pixels on the current image having an amplitude equal to the amplitude of other pixels in the local region may be first screened. If the number of pixels filtered out If the amount exceeds the preset threshold (the threshold can be set as needed), referring to the prior art method, the entire upper image and the entire lower image of the current image are loaded in the memory, and then the selected pixels are judged one by one. Whether it is the point of interest of the current image. On the other hand, if the number of selected pixels does not exceed the preset threshold, then according to the method 1000 shown in FIG. 10, it is determined whether the selected pixels are the points of interest of the current image.

In addition, when determining a point of interest of the current image, the current image may also be decomposed into a plurality of image blocks, wherein vertically adjacent image blocks have at least two rows of overlapping pixels, and horizontally adjacent image blocks have at least two columns overlapping. Pixel. In addition, the size of the image blocks may be the same or different. In fact, each image block is an image, so that the points of interest of each image block can be determined in accordance with the present invention and various methods described in the prior art. For example, when determining the point of interest of each image block, pixels on the current image block with an amplitude equal to the amplitude of other pixels in the local area may be first screened. If the number of the selected pixels exceeds the preset threshold, refer to the prior art method, load the entire upper image block and the entire lower image block of the current image block in the memory, and then judge whether the selected pixels are determined one by one. Is the point of interest of the current image. On the other hand, if the number of selected pixels does not exceed the preset threshold, the method 1000 shown in FIG. 10 is used to judge whether the selected pixels are the points of interest of the current image. After determining the points of interest of each image block, the points of interest of the entire current image can be determined according to the points of interest of each image block, for example, the points of interest of all the image blocks are regarded as the points of interest of the current image.

When the current image is decomposed into a plurality of image blocks, the image block may be represented by the coordinates of the pixel in the upper left corner of the image block and the width and height of the image block, for example, if the current image is 480 wide and 640 high. Dividing into two image blocks of equal size, the two image blocks can be represented in the following manner: The initial pixel coordinates of the first image block are (0,0), the width is 242, and the height is 640. The second image block is initially. The pixel coordinates are (238,0), the width is 242, and the height is 640. If an image with a width of 480 and a height of 640 is divided into four image blocks of equal size, the four image blocks can be represented in the following manner: The initial image coordinates of the first image block are (0, 0) and the width is 242. The height of the second image block is (238,0), the width is 242, and the height is 322. The initial image coordinates of the third image block are (0,318), the width is 242, and the height is 322. Four image block initial pixel coordinates It is (238,318), width is 242, and height is 322.

FIG. 11 is a diagram showing an exemplary hardware configuration of a point of interest judging device 1100 according to an embodiment of the present invention. The point of interest judging means 1100 is configured to determine whether the current pixel is a point of interest of the current image in which the current pixel is located, wherein the current picture is obtained by filtering the target image by using the first filter parameter set. As shown in FIG. 11, the point of interest judging device 1100 includes a memory 1102 and a processor 1104.

The memory 1102 can employ, for example, but not limited to, a Random Access Memory (RAM) or the like. In the point of interest judging device 1100 provided by the embodiment of the present invention, the memory 1102 is configured to store the current image.

The processor 1104 can employ, for example, but not limited to, a general purpose central processor (Central)

Processing Unit (CPU), microprocessor, Application Specific Integrated Circuit (ASIC), etc. In the point of interest judging device 1100 provided by the embodiment of the present invention, the processor 1104 is configured to perform the following operations:

Comparing the amplitude of the current pixel with the amplitude of other pixels in the local area where the current pixel is located on the current image, and determining that the local area is on the target image when the comparison result shows that the amplitude of the current pixel is an extreme value Corresponding area, as the target area;

As described above, in a specific implementation process, each filter parameter group includes a first filter parameter and a second filter parameter, and the filter process includes:

Using the second filter parameter in the filter parameter set used by the filtering process to the first filter map Like performing the second filtering operation, a second filtered image is obtained.

As described above, in a specific implementation process, the first filtering parameter is a Gaussian filtering parameter, the second filtering parameter is a Laplacian filtering parameter, and the first filtering operation is a Gaussian filtering operation, where the The second filtering operation is a Laplacian filtering operation.

As described above, in a specific implementation process, the first filtering parameter in the first filtering parameter group is σ(Ν), and the first filtering parameter in the second filtering parameter group is σ(Ν + 1) The first filter parameter in the third filter parameter set is σ(Ν -1); or

The first filter parameter in the first filter parameter set is _σ (Ν), and the first filter parameter in the second filter parameter set is _σ (Ν -1), in the third filter parameter group The first filter parameter is σ(Ν + 1).

As mentioned above, in the specific implementation process, the following settings can be made, σ(Ν) = 1ί ^Ν σ, where k and j are constants.

As described above, in a specific implementation process, the local area includes at least a current pixel and eight pixels adjacent to the current pixel.

Further details of related technical features (e.g., extreme values, filter parameter sets, etc.) have been described in detail above, and therefore will not be described again here.

It is not difficult to understand that the point of interest judging device 1100 shown in Fig. 11 can be used to implement the point of interest judging method 500 shown in Fig. 5. However, it should be noted that the point of interest judging device 1100 shown in Fig. 11 can also be used to implement the point of interest judging method 900 shown in Fig. 9 and the point of interest judging method 1000 shown in Fig. 10.

Specifically, when the point of interest determination method 900 shown in FIG. 9 is implemented, the point of interest determination apparatus 1100 shown in FIG. 11 is configured to determine whether the current pixel is a point of interest of the current image where the current pixel is located, where the current The image is obtained by filtering the target image by using the first block filter parameter and calculating a Hessian determinant response to the filtered target image, where the current pixel is the response value on the current image. Positive pixel.

When the point of interest determination method 900 shown in FIG. 9 is implemented, the memory 1102 is configured to store the Current image.

When the point of interest determination method 900 shown in FIG. 9 is implemented, the processor 1104 is configured to perform the following operations:

Comparing the amplitude of the current pixel with the amplitude of other pixels in the local region where the current pixel is located on the current image, and determining, when the comparison result shows that the amplitude of the current pixel is a maximum value, determining the local region in the target image The corresponding area on the top, as the target area;

Filtering the target area by using a second block filter parameter, and calculating a Hessian determinant response to the filtered target region to obtain a second region;

Comparing the amplitude of the current pixel with the amplitude of all the pixels in the second region, and when the comparison result shows that the amplitude of the current pixel is a maximum value, filtering the target region by using a third-party frame filtering parameter, and then Calculating a Hessian determinant response to the filtered target region to obtain a third region;

The amplitude of the current pixel is compared with the amplitude of all the pixels in the third region, and when the comparison result shows that the amplitude of the current pixel is a maximum value, it is determined that the current pixel is the point of interest of the current image.

Further details of the relevant technical features (e.g. extreme values, box filtering parameters, etc.) have been described in detail above and will not be described again here.

When the point of interest determination method 1000 shown in FIG. 10 is implemented, the point of interest determination apparatus 1100 shown in FIG. 11 is configured to determine whether the current pixel is a point of interest of the current image in which the current pixel is located, wherein the current image is used. The first filter parameter group is obtained by filtering the target image, and the amplitude of the current pixel is an extreme value compared with the amplitude of other pixels in the local region where the current pixel is located on the current image.

When the point of interest determination method 1000 shown in Fig. 10 is implemented, the memory 1102 is used to store the current image.

When the point of interest determination method 1000 shown in FIG. 10 is implemented, the processor 1104 is configured to perform the following operations:

A corresponding area of the local area on the target image is determined as a target area.

The target area is filtered by using a second filter parameter set to obtain a second area. Comparing the amplitude of the current pixel with the amplitude of all the pixels in the second region, and when the comparison result shows that the amplitude of the current pixel is an extreme value, filtering the target region by using the third filter parameter group to obtain the first Three areas.

It should be noted that although the point of interest judging device 1100 shown in FIG. 11 only shows the memory 1102 and the processor 1104, in the specific implementation process, those skilled in the art should understand that the point of interest judging device 1100 also includes the normal operation. Other devices necessary. At the same time, those skilled in the art will appreciate that the point of interest judging device 1100 may also include hardware devices that implement other additional functions, depending on the particular needs.

FIG. 12 is a diagram showing an exemplary hardware configuration of a point of interest judging device 1200 according to an embodiment of the present invention. The point of interest judging device 1200 is configured to determine whether the current pixel is a point of interest of the current image in which the current pixel is located, wherein the current image is obtained by filtering the target image by using the first filter parameter set. As shown in FIG. 12, the point of interest judging device 1200 includes a memory 1202, a processor 1204, an input/output interface 1206, a communication interface 1208, and a bus 1210.

The functions and implementations of the memory 1202 and the processor 1204 are respectively the same as the memory 1102 and the processor 1104 in the point of interest judging device 1100 described in FIG.

The input/output interface 1206 is for receiving input data and information, and outputting operation results and the like. Communication interface 1208 implements communication between point of interest determination device 1200 and other devices or communication networks using transceivers such as, but not limited to, transceivers.

Bus 1210 can include a path for communicating information between various components of point of interest determining device 1200 (e.g., processor 1202, memory 1204, input/output interface 1206, and communication interface 1208).

FIG. 13 is a schematic diagram showing an exemplary logical structure of a point of interest judging device 1300 according to an embodiment of the present invention. The point of interest judging device 1300 is configured to determine whether the current pixel is located at the current pixel. a point of interest of the front image, wherein the current image is obtained by filtering the target image using the first set of filter parameters. As shown in FIG. 13, the point of interest judging device 1100 includes a main control module 1302, a comparison module 1304, and a filter processing module 1306.

The main control module 1302 is configured to call the comparison module 1304 to compare the amplitude of the current pixel with the amplitude of other pixels in the local area where the current pixel is located on the current image, and the main control module

1302 is further configured to determine, as the target area, a corresponding area of the local area on the target image when the comparison result shows that the amplitude of the current pixel is an extreme value.

The main control module 1302 is further configured to invoke the filter processing module 1306 to perform filtering processing on the target area using the second filter parameter set to obtain a second area.

The main control module 1302 is further configured to call the comparison module 1304 to compare the amplitude of the current pixel with the amplitude of all the pixels in the second region, and the main control module 1302 is further configured to display the amplitude of the current pixel as an extreme value in the comparison result. At the time, the call filter processing module 1306 performs filtering processing on the target area using the third filter parameter set to obtain a third area;

The main control module 1302 is further configured to call the comparison module 1304 to compare the amplitude of the current pixel with the amplitude of all the pixels in the third region, and the main control module 1302 is further configured to display, in the comparison result, that the amplitude of the current pixel is an extreme value. When it is determined, the current pixel is the point of interest of the current image.

As described above, in a specific implementation process, the first filtering parameter in the first filtering parameter group The number is σ(Ν), the first filter parameter in the second filter parameter set is σ(Ν + 1), and the first filter parameter in the third filter parameter set is σ(Ν -1); or

It is not difficult to understand that the point of interest judging device 1300 shown in Fig. 13 can be used to implement the point of interest judging method 500 shown in Fig. 5. However, it should be noted that the point of interest judging device 1300 shown in Fig. 13 can also be used to implement the point of interest judging method 1000 shown in Fig. 10.

Specifically, when the point of interest determination method 1000 shown in FIG. 10 is implemented, the point of interest determination apparatus 1300 shown in FIG. 13 is configured to determine whether the current pixel is a point of interest of the current image where the current pixel is located, where the current The image is obtained by filtering the target image using the first filter parameter set, and the amplitude of the current pixel is an extreme value compared with the amplitude of other pixels in the local region where the current pixel is located on the current image.

When the point of interest determination method 1000 shown in FIG. 10 is implemented, the main control module 1302 is configured to determine a corresponding area of the local area on the target image as a target area.

The main control module 1302 is further configured to compare the amplitude of the current pixel with the amplitude of all the pixels in the second area, and the main control module 1302 is further configured to: when the comparison result shows that the amplitude of the current pixel is an extreme value Calling the filter processing module 1306 to use the third filter parameter set Filtering the target area to obtain a third area.

The main control module 1302 is further configured to compare the amplitude of the current pixel with the amplitude of all the pixels in the third region, and the main control module 1302 is further configured to display, when the comparison result shows that the amplitude of the current pixel is an extreme value. , determining that the current pixel is a point of interest of the current image.

The contents of the first filter parameter group, the second filter parameter group, the third filter parameter group, the extreme value, the local region, the filter processing, the corresponding region, and the like have been described in detail in the foregoing, and therefore will not be described herein. .

FIG. 14 is a diagram showing an exemplary logical structure of a point of interest judging device 1400 according to an embodiment of the present invention. The point of interest judging device 1400 is configured to determine whether the current pixel is a point of interest of the current image where the current pixel is located, wherein the current image is a filter target processed by using the first block filter parameter and then filtered. The image is obtained by calculating a Hessian determinant response, and the current pixel is a pixel whose positive response value is on the current image. As shown in FIG. 14, the point of interest judging device 1400 includes a main control module 1402, a comparison module 1404, a filtering processing module 1406, and a calculation module 1408.

The main control module 1402 is configured to compare the amplitude of the current pixel with the amplitude of other pixels in the local area where the current pixel is located on the current image, and the main control module 1402 is further configured to display the amplitude of the current pixel in the comparison result. When it is a maximum value, a corresponding area of the local area on the target image is determined as a target area.

The main control module 1402 is further configured to call the filter processing module 1406 to filter the target area by using the second block filter parameter, and then call the calculation module 1408 to calculate a Hessian determinant response to the filtered target region to obtain a second region. ;

The main control module 1402 is further configured to compare the amplitude of the current pixel with the amplitude of all the pixels in the second region, and the main control module 1402 is further configured to display the amplitude of the current pixel as a maximum value in the comparison result. The call filter processing module 1406 performs filtering processing on the target area using the third-party frame filter parameter, and then calls the calculation module 1408 to calculate a Hessian determinant response to the filtered target region to obtain a third region;

The main control module 1402 is further configured to call the comparison module 1404 to compare the amplitude of the current pixel with the first The main control module 1402 is further configured to determine that the current pixel is a point of interest of the current image when the comparison result shows that the amplitude of the current pixel is a maximum value.

It is not difficult to understand that the point of interest judging device 1400 shown in Fig. 14 can be used to implement the point of interest judging method 900 shown in Fig. 9.

A person skilled in the art may know that all or part of the above steps may be completed by hardware related to program instructions, and the program may be stored in a computer readable storage medium such as a ROM, a RAM, an optical disc, or the like. .

In conclusion, the above is only a preferred embodiment of the present invention and is not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and scope of the invention are intended to be included within the scope of the invention.

Claims

claims

1. An interest point determination method, used to determine whether the current pixel is an interest point of the current image where the current pixel is located, wherein the current image is obtained by filtering the target image using a first filter parameter group, It is characterized in that the method includes:

Compare the amplitude of the current pixel with the amplitude of other pixels in the local area where the current pixel is located on the current image. When the comparison result shows that the amplitude of the current pixel is an extreme value, determine that the local area is on the target image. The corresponding area of is used as the target area;

Use the second filter parameter group to filter the target area to obtain the second area; Compare the amplitude of the current pixel with the amplitude of all pixels in the second area, and the comparison result shows that the amplitude of the current pixel is one pole value, use the third filtering parameter group to perform filtering processing on the target area to obtain the third area;

The amplitude of the current pixel is compared with the amplitudes of all pixels in the third area. When the comparison result shows that the amplitude of the current pixel is an extreme value, the current pixel is determined to be the interest point of the current image.

2. The method of claim 1, wherein each filtering parameter group includes a first filtering parameter and a second filtering parameter, and the filtering process includes:

Perform a first filtering operation on the image to be processed using the first filtering parameter in the filtering parameter group used in the filtering process to obtain the first filtered image;

Perform a second filtering operation on the first filtered image using the second filtering parameter in the filtering parameter group used in the filtering process to obtain a second filtered image.

3. The method of claim 2, wherein the first filtering parameter is a Gaussian filtering parameter, the second filtering parameter is a Laplacian filtering parameter, and the first filtering operation is a Gaussian filtering operation. , the second filtering operation is a Laplacian filtering operation.

4. The method of claim 2, wherein the first filter parameter in the first filter parameter group is σ(N), and the first filter parameter in the second filter parameter group is σ(N) N + 1), the first filter parameter in the third filter parameter group is σ (N -1); or

The first filter parameter in the first filter parameter group is _σ (N), and the second filter parameter The first filter parameter in the group is _σ (N -1), and the first filter parameter in the third filter parameter group is σ (N + 1).

5. The method of claim 4, wherein _CT (N) = k ^N _CT , where k and j are constants.

6. The method of claim 1, wherein the local area includes at least a current pixel and 8 pixels adjacent to the current pixel.

7. An interest point judgment device, used to judge whether the current pixel is an interest point of the current image where the current pixel is located, wherein the current image is obtained by filtering the target image using a first filter parameter group, It is characterized in that the device includes:

Memory, used to store the current image;

Processor, used to perform the following operations:

Compare the amplitude of the current pixel with the amplitude of other pixels in the local area where the current pixel is located on the current image. When the comparison result shows that the amplitude of the current pixel is an extreme value, determine that the local area is on the target image. The corresponding area of , as the target area; Use the second filtering parameter group to filter the target area to obtain the second area;

Compare the amplitude of the current pixel with the amplitude of all pixels in the second area. When the comparison result shows that the amplitude of the current pixel is an extreme value, use the third filter parameter group to filter the target area to obtain the third Three areas;

The amplitude of the current pixel is compared with the amplitudes of all pixels in the third area. When the comparison result shows that the amplitude of the current pixel is an extreme value, the current pixel is determined to be the point of interest of the current image.

8. The device according to claim 7, wherein each filtering parameter group includes a first filtering parameter and a second filtering parameter, and the filtering process includes:

Perform a first filtering operation on the image to be processed using the first filtering parameter in the filtering parameter group used in the filtering process to obtain the first filtered image; Perform a second filtering operation on the first filtered image using the second filtering parameter in the filtering parameter group used in the filtering process to obtain a second filtered image.

9. The device of claim 8, wherein the first filtering parameter is a Gaussian filtering parameter, the second filtering parameter is a Laplacian filtering parameter, and the first filtering operation is a Gaussian filtering operation. , the second filtering operation is a Laplacian filtering operation.

10. The device according to claim 8, wherein the first filter parameter in the first filter parameter group is σ(N), and the first filter parameter in the second filter parameter group is σ(N) N + 1), the first filter parameter in the third filter parameter group is _σ ( N - 1); or

The first filter parameter in the first filter parameter group is _σ (N), the first filter parameter in the second filter parameter group is σ (N -1), and the third filter parameter group The first filtering parameter is σ(N + 1).

11. The device of claim 10, wherein _CT (N) = k ^N _CT , where k and j are constants.

12. The device according to claim 7, wherein the local area includes at least a current pixel and 8 pixels adjacent to the current pixel.