Extraction method of feature point descriptors for weak texture representation

Abstract

The present application discloses a method, apparatus, device, and storage medium for extracting feature point descriptors for weak texture representation. The method comprises: inputting a liver image of a preset size; processing the liver image through a Gaussian filter unit and a filter unit to generate a feature descriptor; and outputting the feature descriptor and the corresponding corner point coordinates. The above-mentioned solution provided by the present application can extract more effective feature point pairs and, based on additional information such as the extracted angle, size, and distance, add a filter layer to effectively filter out invalid feature point pairs. At the same time, the present invention adopts accelerated computing on a PYNQ board, and the result can reach 60 frames / s, meeting the real-time requirements of the system.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a method, device, equipment and storage medium for extracting feature point descriptors for weak texture representation. Background Art

[0002] With the prevalence of robotic-assisted surgery and the development of computer vision technology, the domestic market for medical endoscopy equipment is booming. Minimally invasive surgery (MIS) is becoming an increasingly important technique in daily medical practice. Minimally invasive surgery has gained widespread application in recent years due to its small incisions, minimal pain, and rapid postoperative recovery. However, while benefiting patients, minimally invasive surgery also places higher demands on the surgeon. Minimally invasive surgery involves multi-person indirect eye-hand collaboration using an endoscope as a medium. Laparoscopic technology is widely used in minimally invasive surgery. During surgery, an endoscope is used to obtain endoscopic image information and visualize the surface of organs within the body, achieving excellent results while ensuring surgical safety. However, the limited field of view of the endoscope and the lack of direct depth perception make it difficult for surgeons to position themselves during surgery, significantly impacting the accuracy and safety of surgical procedures. Higher-quality 3D point cloud images provide better depth information. Extracting image feature points is fundamental image processing for 3D reconstruction. Image processing and pattern classification are at the core of computer vision applications. Key steps include detecting image feature points, generating feature descriptors for these points, and matching these descriptors. Feature point detection algorithms decompose an image into points of interest and irrelevant information. The process of mathematically describing and memorizing the information surrounding these points is called feature point description.

[0003] Because most image processing algorithms are computationally intensive and complex, meeting real-time requirements on traditional PC platforms is difficult. Therefore, porting and redeveloping image processing algorithms to embedded devices has become a key breakthrough. Embedded devices suitable for porting image processing algorithms include FPGAs (Field Programmable Gate Arrays), GPUs (Graphics Processing Units), and ASICs (Application-Specific Integrated Circuits), each with its own advantages. Embedded hardware acceleration utilizes specialized hardware devices, rather than CPUs, to handle computationally intensive tasks. This fully exploits the hardware's parallel computing capabilities, enabling deployment at the edge without being constrained by size or battery life. GPUs were initially used to accelerate image display tasks. Later, researchers discovered that GPUs are suitable for parallel processing of large amounts of data, embodying their high bandwidth, high clock speed, and high parallelism. GPUs excel at handling multiple tasks, especially repetitive computations, such as graphics and deep learning. GPUs allow for the design of parallel data processing structures, offer higher floating-point throughput and processing power, and possess significantly greater memory bandwidth than CPU processors. These advantages make GPUs the primary choice for high-performance computing. However, using GPU-based accelerators is power inefficient. Another alternative accelerator is the FPGA. Current FPGAs offer a good balance between processing speed and power consumption. More importantly, FPGAs provide data flow constraints and can process data in parallel. These advantages make FPGAs competitive with GPUs. However, in a complete image processing system, a single embedded device struggles to meet the tight integration of compute-intensive and control-intensive algorithms. Most current algorithm processing systems are heterogeneous. In these heterogeneous systems, FPGAs and GPUs accelerate image processing algorithms, while the CPU handles overall control and scheduling.

[0004] Accelerating descriptor matching systems in hardware is a breakthrough for improving the performance of today's high-speed vision devices. In recent years, numerous researchers have focused on reducing the dimensionality of descriptors while maintaining performance, thereby reducing the complexity of descriptor computation and accelerating computation. The BRIEF descriptor generation algorithm has been proposed in the prior art. The BRIEF algorithm, centered around a detected feature point, compares pixel intensities by selecting points that conform to a Gaussian distribution. The comparison result is recorded as a binary string, which is called the BRIEF descriptor for that feature point. The Hamming distance between two binary BRIEF descriptors can be calculated to determine whether the descriptors match, simplifying the descriptor matching process and accelerating it. Furthermore, the existing feature point description algorithm, FREAK (Fast Retina Keypoint), is a further improvement on the BRISK algorithm's sampling model. Its sampling pattern mimics the structure of the human retina, using a coarse-to-fine process that reduces computational complexity and outperforms SIFT and SURF algorithms in terms of speed. Based on the ORB corner detection and FREAK descriptor algorithms, by replacing the Harris evaluation criterion in the ORB algorithm with the Hessian evaluation criterion and modifying the sampling mode on FREAK, it is ultimately possible to screen out accurately matching feature points at each layer of the size pyramid and perform computational acceleration on an FPGA.

[0005] To obtain high-quality 3D point cloud images, a large number of endoscopic image sequences must be processed, which is computationally prohibitive for traditional PCs. Consequently, more and more researchers are attempting to port image processing algorithms to various embedded devices for secondary development. Embedded devices suitable for image processing algorithm porting include FPGAs (Field Programmable Gate Arrays), GPUs (Graphics Processing Units), and ASICs (Application-Specific Integrated Circuits), each with its own advantages. Furthermore, endoscopic image acquisition is limited by the fact that a cold light source is attached to the top of the endoscope for image acquisition. Variations in organ surface brightness and surface smoothness during feature extraction can affect feature extraction and ultimately impact the back-end 3D reconstruction results.

[0006] Computer vision researchers have developed numerous stable local image features over the years, including the well-known SIFT, SURF, and ORB. These manually designed features vary in distinctiveness, efficiency, and reusability. They are particularly effective for endoscopic images, which have low texture. Summary of the Invention

[0007] In view of the above-mentioned defects or deficiencies in the prior art, it is desirable to provide a method, apparatus, device and storage medium for extracting feature point descriptors for weak texture representation.

[0008] In a first aspect, an embodiment of the present application provides a method for extracting feature point descriptors for weak texture representation, the method comprising: inputting a liver image of a preset size; processing the liver image through a Gaussian filter unit and a Filter unit to generate a feature descriptor; and outputting the feature descriptor and the corresponding corner point coordinates.

[0009] In one embodiment, the processing of the liver image through a Gaussian filter unit includes: inputting the liver image into the Gaussian filter unit; the Gaussian filter unit is used to perform Gaussian blur processing on the liver image to remove noise in the image and generate a smooth image.

[0010] In one embodiment, the Filter unit processing includes: sequentially processing the liver image through a FAST feature point extraction unit, a non-maximum suppression unit, a multi-neighborhood feature information extraction unit, and a maximum corner point suppression unit.

[0011] In one embodiment, generating the feature descriptor includes: modeling the relationship between different liver image neighborhoods; and obtaining the feature descriptor of the feature point according to the feature value in the corresponding model.

[0012] In one embodiment, after generating the feature descriptor, the method further includes: sorting the feature points and grouping the feature points in order; eliminating feature points that do not meet the conditions; calculating the angles, sizes, and distances of a preset number of feature points; inputting constraint conditions, and when all the constraint conditions are met, determining that the matching structure is correct, otherwise removing the corresponding feature point pairs.

[0013] In one embodiment, the processing of the liver image through a Gaussian filter unit and a Filter unit includes: processing the liver image through a Gaussian filter unit and a Filter unit in a PYNQ processing platform, wherein the PYNQ processing platform includes an ARM processor and an FPGA hardware acceleration module.

[0014] In a second aspect, an embodiment of the present application also provides a device for extracting feature point descriptors for weak texture representation, the device comprising: an input unit for inputting a liver image of a preset size; a processing unit for processing the liver image through a Gaussian filter unit and a Filter unit to generate a feature descriptor; and an output unit for outputting the feature descriptor and the corresponding corner point coordinates.

[0015] In a third aspect, an embodiment of the present application further provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements any of the methods described in the embodiments of the present application.

[0016] In a fourth aspect, an embodiment of the present application further provides a computer device and a computer-readable storage medium on which a computer program is stored, wherein the computer program is used to implement any method described in the embodiments of the present application when the computer program is executed by a processor.

[0017] Beneficial effects of the present invention:

[0018] The method for extracting feature point descriptors for weak texture representations, provided by this invention, utilizes a specific feature descriptor designed specifically for the limited texture detail of liver surface images. This allows for the extraction of a large number of valid feature point pairs. Furthermore, a filter layer is added based on additional information such as angle, size, and distance extracted to effectively filter out invalid feature point pairs. Furthermore, the method utilizes accelerated computing on a PYNQ board, achieving a frame rate of 60 frames per second, meeting the system's real-time requirements. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Other features, objects and advantages of the present application will become more apparent upon reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0020] Figure 1 A schematic flow chart of a method for extracting feature point descriptors for weak texture representation provided by an embodiment of the present application is shown;

[0021] Figure 2 FIG2 shows an exemplary structural block diagram of an apparatus 200 for extracting feature point descriptors for weak texture representation according to an embodiment of the present application;

[0022] Figure 3 A schematic diagram showing the structure of a computer system suitable for implementing a terminal device according to an embodiment of the present application is shown;

[0023] Figure 4 A schematic diagram of the algorithm IP module provided in an embodiment of the present application is shown;

[0024] Figure 5 A schematic diagram of the algorithm flow provided in the embodiment of the present application is shown;

[0025] Figure 6 A schematic diagram of multiple neighborhoods provided by an embodiment of the present application is shown;

[0026] Figure 7 A schematic diagram of parallel computing provided by an embodiment of the present application is shown. DETAILED DESCRIPTION

[0027] To make the above-mentioned objects, features, and advantages of the present invention more readily apparent, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings. The following description sets forth numerous specific details to facilitate a full understanding of the present invention. However, the present invention can be implemented in many other ways than those described herein, and those skilled in the art may make similar modifications without departing from the scope of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0028] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise", "axial", "radial", "circumferential" and the like to indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0029] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of the technical features being referred to. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "plurality" means at least two, such as two, three, etc., unless otherwise specifically defined.

[0030] In the present invention, unless otherwise specified or limited, the terms "installed," "connected," "connect," "fixed," etc. should be understood in a broad sense. For example, they can refer to fixed connection, detachable connection, or integration; mechanical connection, electrical connection; direct connection, or indirect connection through an intermediate medium; internal communication between two components, or interaction between two components, unless otherwise specified. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0031] In the present invention, unless otherwise expressly specified or limited, when a first feature is "above" or "below" a second feature, it may mean that the first and second features are in direct contact, or that the first and second features are in indirect contact through an intermediary. Furthermore, when a first feature is "above," "above," or "above" a second feature, it may mean that the first feature is directly above or diagonally above the second feature, or simply means that the first feature is at a higher level than the second feature. When a first feature is "below," "below," or "below" a second feature, it may mean that the first feature is directly below or diagonally below the second feature, or simply means that the first feature is at a lower level than the second feature.

[0032] It should be noted that when an element is referred to as being "fixed to" or "disposed on" another element, it may be directly on the other element or there may be an intermediate element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be an intermediate element. The terms "vertical," "horizontal," "upper," "lower," "left," "right," and similar expressions used herein are for illustrative purposes only and do not represent the only implementation methods.

[0033] Please refer to Figure 1 , Figure 1 A flow chart of a method for extracting feature point descriptors for weak texture representation provided by an embodiment of the present application is shown.

[0034] like Figure 1 As shown, the method includes:

[0035] Step 110, inputting a liver image of a preset size;

[0036] Step 120, processing the liver image through a Gaussian filter unit and a Filter unit to generate a feature descriptor;

[0037] Step 130: Output feature descriptors and corresponding corner point coordinates.

[0038] Using this technical solution, a specific feature descriptor was designed to address the low texture detail of liver surface images. This allows for the extraction of a large number of valid feature point pairs. Furthermore, a filter layer was added based on additional information such as angle, size, and distance, effectively filtering out invalid feature point pairs. Furthermore, the present invention utilizes accelerated computing on a PYNQ board, achieving a frame rate of 60 frames per second, meeting the system's real-time requirements.

[0039] In some embodiments, the processing of the liver image by a Gaussian filter unit in the present application includes: inputting the liver image into the Gaussian filter unit; the Gaussian filter unit is used to perform Gaussian blur processing on the liver image to remove noise in the image and generate a smooth image.

[0040] In some embodiments, the Filter unit processing in the present application includes: processing the liver image in sequence through a FAST feature point extraction unit, a non-maximum suppression unit, a multi-neighborhood feature information extraction unit, and a maximum corner point suppression unit.

[0041] In some embodiments, generating feature descriptors in the present application includes: modeling the relationship between different liver image neighborhoods; and obtaining feature descriptors of feature points according to feature values ​​in the corresponding model.

[0042] In some embodiments, after generating feature descriptors in the present application, the method further includes: sorting the feature points and grouping the feature points in order; eliminating feature points that do not meet the conditions; calculating the angles, sizes, and distances of a preset number of feature points; inputting constraints, and when all constraints are met, the matching structure is determined to be correct, otherwise the corresponding feature point pairs are removed.

[0043] In some embodiments, the liver image is processed by a Gaussian filter unit and a Filter unit in the present application, including: the liver image is processed by a Gaussian filter unit and a Filter unit in a PYNQ processing platform, and the PYNQ processing platform includes an ARM processor and an FPGA hardware acceleration module.

[0044] This paper mainly designs a feature descriptor for liver images that records more information. In addition, a filter layer is added to the extra information, which is used to pass through the point pairs classified in the early extraction stage. Due to the large amount of calculation, it was finally decided to accelerate the operation of the feature extraction and matching algorithm modules on the PYNQ board. Finally, the various algorithm submodules are integrated into various IPs through the HLS tool, such as Figure 4 As shown in the figure, this feature descriptor incorporates the concept of multi-neighborhoods. Compared to other feature point algorithms, it can extract more feature points and has high discriminability and strong robustness. Based on this additional local image information, a filter layer is added to filter out mismatched feature point pairs. It can still maintain high matching accuracy even for images with little texture and large rotation angles.

[0045] The traditional SLAM framework is mainly divided into two modules: front-end visual odometry and back-end optimization. The front-end visual odometry mainly includes the extraction of special diagnosis points and the matching of feature descriptors, which are accelerated by parallel calculations on the PYNQ board. The specific algorithm modules are as follows: Figure 5 shown.

[0046] In ORB_SLAM, it is known for its fast speed by mainly detecting places where local pixel changes are obvious in grayscale images. For a pixel p, a common method is to select N pixels on a circle with a radius of m. If the brightness of the selected N points is greater than Ip+T or less than Ip-T,n, then the pixel p can be considered as a feature point. However, for laparoscopic liver images, there is little surface texture, and only a small number of feature points can be extracted using this method. In order to extract enough effective feature points in scenes with weak texture or no texture, the present invention introduces the concept of multi-neighborhood to describe the gradient changes of image feature points.

[0047] by Figure 6 Taking the multi-neighborhood method as an example, based on the concept of the multi-neighborhood method, the present invention uses a series of neighborhoods I = [I1, I2, ..., Ii, ... Im] eigenvalues ​​and eigenvectors as the neighborhood feature descriptors of the pixel p. Assume that p and q marked in the figure are the feature points to be detected. There can be i circular neighborhoods I1 and I2 of different sizes centered on p and q, respectively, where the diameter of the neighborhood I1 is smaller than the diameter of the neighborhood I2. When the pixel information in the neighborhood I1 of point p and point q is the same, the line segments where points p and q are located can be rotated to overlap with each other. At this time, neither the algorithm nor the human eye can distinguish between the two feature points. However, simply expanding the neighborhood range of the feature point cannot express the structural information of the image and may introduce error information. Therefore, the multi-neighborhood method is adopted here to obtain the feature descriptors of the feature points by modeling the relationship between different neighborhoods, thereby realizing the expression of image structural information and texture information.

[0048] The most important feature of an image feature point is the point where the grayscale value changes within a certain range. For a pixel point p, the present invention defines its gradient in the x-direction and y-direction as g x and g y .

[0049] (Formula 1)

[0050] (Formula 2)

[0051] According to the distribution of gradients in the neighborhood of feature points, the local tensor of the image is defined as

[0052] (Formula 3)

[0053] in: The standard deviation is Gaussian function, and are the horizontal gradient and vertical gradient of pixel p respectively. is symmetric and positive semidefinite, so it has two orthogonal eigenvectors:

[0054] (Formula 4)

[0055] (Formula 5)

[0056] The corresponding eigenvalues ​​are

[0057] (Formula 6)

[0058] (Formula 7)

[0059] The larger the eigenvalue, the greater the proportion of the image gradient distribution in the direction of the corresponding eigenvector. > ,so is the main eigenvalue, then The main eigenvector is the eigenvector of the feature point neighborhood. The rotation invariance of the feature descriptor is further realized based on the eigenvector of the feature point neighborhood.

[0060] After extracting and matching ORB features, the present invention applies the improved F-SORT algorithm to further improve the matching accuracy. First, the present invention sorts the feature points and groups them in order. For a given matching feature point, in image I A The characteristic point in is P={p i}, and assume that in image I B The corresponding feature point is Q={q pi}, and the present invention defines this pair of mutually matching feature points as M = {m i =(p i , q pi )}, then the present invention stores the feature points in ascending order according to the x-coordinates of the image feature points, and the present invention defines the sorted and stored feature points as { }, for each feature point , the present invention corresponds to the corresponding matching feature point as Q, and is defined as Then, the present invention divides the matched feature points into k groups using the following formula.

[0061] (Formula 8)

[0062] in After the feature points are classified, the present invention removes the feature point pairs that do not meet the conditions. This is only the first step of filtering the feature point pairs in the present invention.

[0063] Then, the present invention calculates the angle, size, and distance of the k feature point pairs. This is the second step of the present invention in filtering the feature point pairs.

[0064] (Formula 9)

[0065] (Formula 10)

[0066] (Formula 11)

[0067] in and s are the angle and size of the matching feature point pair, and is a pair of feature points (p i ,p j ). Therefore, the present invention defines a constraint condition as follows

[0068] (Formula 12)

[0069] in For the matching feature point pairs Consistency constraints, and sum(.) represents the addition operation. Med(.) represents the operation of taking the middle value, and and When all the constraints are met, the present invention considers that the matching result is correct, otherwise, the present invention removes the pair of feature points.

[0070] PYNQ is a heterogeneous system with an ARM processor as the main controller and an FPGA as the accelerator module. Therefore, parallelization mechanisms are crucial for improving computational throughput. The parallelized pipeline used is shown in the figure. For normal frame processing, while the ARM processor performs pose estimation and optimization, the ORB Extractor and Brief Matcher modules have already begun feature point extraction and feature descriptor matching for the next frame. However, keyframe processing is different because the map update is performed on the ARM processor after pose estimation and optimization are complete. The ORB Extractor module performs feature extraction on the FPGA in parallel with the ARM processor, but the Brief Matcher module begins its work after the map update is complete. Through the above parallelization mechanism, several stages can be efficiently executed in the pipeline. For normal frames, feature extraction and matching run in parallel with pose estimation and optimization. For keyframes, feature extraction runs in parallel with pose estimation and optimization. This parallel processing significantly improves computational throughput.

[0071] refer to Figure 7 As shown, Figure 7Here, FE represents the feature extraction module, FM represents the feature descriptor matching module, PE represents the pose estimation module, and PO represents the pose optimization module and map update module. This parallelization mechanism allows for efficient execution of multiple stages within the pipeline. For normal frames, feature extraction and matching run in parallel with pose estimation and optimization. For key frames, feature extraction runs in parallel with pose estimation and optimization. This parallel processing significantly improves computational throughput.

[0072] Further, refer to Figure 2 , Figure 2 FIG2 shows an exemplary structural block diagram of an apparatus 200 for extracting feature point descriptors for weak texture representation according to an embodiment of the present application.

[0073] like Figure 2 As shown, the device includes:

[0074] An input unit 210 is used to input a liver image of a preset size;

[0075] a processing unit 220, configured to generate a feature descriptor after processing the liver image through a Gaussian filter unit and a Filter unit;

[0076] The output unit 230 is configured to output feature descriptors and corresponding corner point coordinates.

[0077] It should be understood that the units or modules described in the apparatus 200 are similar to those described in the reference Figure 1 The various steps in the described method correspond to each other. Therefore, the operations and features described above for the method are also applicable to the device 200 and the units contained therein, and will not be repeated here. The device 200 can be pre-implemented in the browser or other security application of the electronic device, or loaded into the browser or its security application of the electronic device by downloading or other means. The corresponding units in the device 200 can cooperate with the units in the electronic device to implement the solution of the embodiment of the present application.

[0078] Reference below Figure 3 , which shows a structural diagram of a computer system 300 suitable for implementing a terminal device or server of an embodiment of the present application.

[0079] like Figure 3As shown, computer system 300 includes a central processing unit (CPU) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage unit 308 into a random access memory (RAM) 303. Various programs and data required for the operation of system 300 are also stored in RAM 303. CPU 301, ROM 302, and RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to bus 304.

[0080] The following components are connected to the I / O interface 305: an input section 306 including a keyboard, mouse, and the like; an output section 307 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and speakers; a storage section 308 including devices such as a hard disk; and a communication section 309 including a network interface card such as a LAN card or a modem. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to the I / O interface 305 as needed. Removable media 311, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 310 as needed, so that computer programs read from the media can be installed in the storage section 308 as needed.

[0081] In particular, according to the embodiments of the present disclosure, the above reference Figure 1 The described process can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a method for extracting feature point descriptors for weak texture representation, which includes a computer program tangibly embodied on a machine-readable medium, the computer program including a method for executing Figure 1 In such an embodiment, the computer program may be downloaded and installed from a network via the communication section 309 and / or installed from the removable medium 311 .

[0082] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the aforementioned module, program segment, or a part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0083] The units or modules involved in the embodiments described in the present application can be implemented by software or by hardware. The described units or modules can also be set in a processor. For example, they can be described as: a processor includes a first sub-area generation unit, a second sub-area generation unit, and a display area generation unit. Among them, the names of these units or modules do not constitute a limitation of the unit or module itself under certain circumstances. For example, the display area generation unit can also be described as "a unit for generating a display area for text based on the first sub-area and the second sub-area".

[0084] As another aspect, the present application further provides a computer-readable storage medium, which may be the computer-readable storage medium included in the aforementioned apparatus in the above-mentioned embodiment, or may be a separate computer-readable storage medium not incorporated into the apparatus. The computer-readable storage medium stores one or more programs, which are used by one or more processors to execute the text generation method for a transparent window envelope described in the present application.

[0085] The above description is merely a preferred embodiment of the present application and an illustration of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to the technical solutions formed by the specific combination of the above-mentioned technical features, but also encompasses other technical solutions formed by any combination of the above-mentioned technical features or their equivalents without departing from the aforementioned inventive concept. For example, a technical solution formed by replacing the above-mentioned features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for extracting feature point descriptors for weak texture representation, characterized in that: The method includes: Input liver image of preset size; Processing the liver image through a Gaussian filter unit and a Filter unit in a PYNQ processing platform to generate a feature descriptor, wherein the PYNQ processing platform includes an ARM processor and an FPGA hardware acceleration module; The Filter unit processing includes: The liver image is processed sequentially by a FAST feature point extraction unit, a non-maximum suppression unit, a multi-neighborhood feature information extraction unit, and a maximum corner point suppression unit; The generating of the feature descriptor comprises: Modeling the relationships between different liver image neighborhoods; Obtain the feature descriptor of the feature point according to the feature value in the corresponding model, specifically including: Use a series of neighborhood eigenvalues ​​and eigenvectors as the neighborhood feature descriptors of pixel p; Define pixel point p, the gradient in the x and y directions is g x and g y : According to the distribution of gradients in the neighborhood of feature points, the local tensor of the image is defined as: in, The standard deviation is Gaussian function, and are the horizontal gradient and vertical gradient of pixel p, respectively. is symmetric and positive semidefinite, with two orthogonal eigenvectors: The corresponding eigenvalues ​​are: in, > , is the main eigenvalue, is the main eigenvector; Sort the feature points and group them in order; Eliminate feature points that do not meet the conditions; Calculate the angle, size, and distance of a preset number of feature point pairs; Input constraints. When all constraints are met, the matching structure is determined to be correct. Otherwise, the corresponding feature point pairs are removed. Output feature descriptors and corresponding corner coordinates.

2. The method for extracting feature point descriptors for weak texture representation according to claim 1, characterized in that: The processing of the liver image by the Gaussian filter unit comprises: inputting the liver image into the Gaussian filtering unit; The Gaussian filter unit is used to perform Gaussian blur processing on the liver image to remove noise in the image and generate a smooth image.

3. A device for extracting feature point descriptors for weak texture representation, characterized in that: The device includes: An input unit, used for inputting a liver image of a preset size; A processing unit is configured to generate a feature descriptor after the liver image is processed by a Gaussian filter unit and a Filter unit in a PYNQ processing platform, wherein the PYNQ processing platform includes an ARM processor and an FPGA hardware acceleration module; the Filter unit processing includes: sequentially processing the liver image by a FAST feature point extraction unit, a non-maximum suppression unit, a multi-neighborhood feature information extraction unit, and a maximum corner suppression unit; generating the feature descriptor includes: modeling the relationship between different liver image neighborhoods; obtaining a feature descriptor of the feature point based on the eigenvalues ​​in the corresponding model, specifically including: using a series of neighborhood eigenvalues ​​and eigenvectors as neighborhood feature descriptors of pixel p; defining the gradient of pixel p in the x-direction and y-direction as g x and g y : According to the distribution of gradients in the neighborhood of feature points, the local tensor of the image is defined as: in, The standard deviation is Gaussian function, and are the horizontal gradient and vertical gradient of pixel p, respectively. is symmetric and positive semidefinite, with two orthogonal eigenvectors: The corresponding eigenvalues ​​are: in, > , is the main eigenvalue, The main feature vector is used; the feature points are sorted and grouped in order; feature points that do not meet the conditions are eliminated; the angles, sizes, and distances of a preset number of feature point pairs are calculated; the constraints are input, and when all the constraints are met, the matching structure is determined to be correct, otherwise the corresponding feature point pairs are removed; The output unit is used to output feature descriptors and corresponding corner point coordinates.

4. The device for extracting feature point descriptors for weak texture representation according to claim 3, characterized in that: The processing of the liver image by the Gaussian filter unit comprises: inputting the liver image into the Gaussian filtering unit; The Gaussian filter unit is used to perform Gaussian blur processing on the liver image to remove noise in the image and generate a smooth image.

5. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the program, the method according to any one of claims 1 to 2 is implemented.

6. A computer-readable storage medium having stored thereon a computer program for: When the computer program is executed by a processor, the method according to any one of claims 1 to 2 is implemented.

Images