Game icon positioning method and system based on feature vectors

Abstract

The invention provides a game icon positioning method and system based on feature vectors, and relates to the technical field of image processing, and the method comprises the steps: obtaining a to-be-processed game icon; executing a sliding window operation on the pre-processed game icon to generate a plurality of candidate areas; extracting a global feature vector and a local feature vector of the candidate region through an XFeat deep neural network; standardizing the global feature vector and the local feature vector to obtain a standardized feature vector; performing quality similarity matching on the standardized feature vector and a template feature vector in an icon template library; performing post-processing on the matching result to obtain an optimized matching result; and outputting position information and category information of the game icons according to an optimized matching result to complete positioning of the game icons. Through feature vector extraction and rapid retrieval, high-robustness, high-efficiency and high-accuracy matching and positioning of game icons are realized, and dynamic expansion of a template library is supported.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and system for locating game icons based on feature vectors. Background Technology

[0002] Existing methods for game interface icon recognition largely rely on traditional template matching or object detection algorithms, which suffer from sensitivity to icon size, rotation, occlusion, and background changes, as well as low matching accuracy. Furthermore, template libraries have poor scalability; adding new icons requires retraining the entire system, and they are inefficient for scenarios with high real-time matching requirements. Therefore, the industry urgently needs a novel icon matching and localization method that achieves high robustness, high efficiency, and scalability.

[0003] Current methods for recognizing game interface icons mostly rely on traditional template matching or object detection algorithms. By pre-setting icon templates or training detection models, the icons of characters, equipment, and items in the game interface are identified and located. In early, simple game interface scenarios, these methods were used to implement basic icon matching functions.

[0004] However, existing technologies are sensitive to icon size, rotation, occlusion, and background changes, resulting in low matching accuracy. Furthermore, the template library has poor scalability, requiring retraining for new icons. At the same time, traditional methods rely solely on cosine similarity for decision-making, lacking an effective quality assessment mechanism, leading to insufficient positioning accuracy under complex game screens and interference from similar icons. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the purpose of this invention is to provide a game icon localization method based on feature vectors, which can solve the technical problems of low matching accuracy, poor template library scalability, and insufficient localization accuracy under complex game screens and interference from similar icons.

[0006] A first aspect of this invention proposes a game icon localization method based on feature vectors, comprising: S1: Get the game icon to be processed.

[0007] S2: Preprocess the game icon.

[0008] S3: Perform a sliding window operation on the preprocessed game icon to generate multiple candidate regions.

[0009] S4: Extract global and local feature vectors of candidate regions using the XFeat deep neural network.

[0010] S5: Standardize the global and local feature vectors to obtain standardized feature vectors.

[0011] S6: Perform quality similarity matching between the standardized feature vector and the template feature vector in the icon template library.

[0012] S7: Perform post-processing on the matching results, including nonmaximum suppression, multi-scale fusion, and confidence filtering, to obtain optimized matching results.

[0013] S8: Based on the optimized matching results, output the location and category information of the game icon to complete the positioning of the game icon.

[0014] A second aspect of this invention provides a game icon localization system based on feature vectors, comprising: a processor and a memory; The memory stores programs or instructions that can run on the processor, which, when executed by the processor, implement the steps of the feature vector-based game icon localization method as described in the first aspect.

[0015] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following: In this embodiment of the invention, robust global and local features are extracted using an XFeat deep neural network, and multi-dimensional data augmentation is combined to effectively overcome the sensitivity issues caused by icon scale, rotation, and background changes. To address the template library scalability issue, an efficient vector index based on FAISS is constructed with a dynamic update mechanism, enabling rapid expansion of the template library without retraining. Furthermore, a QATM quality-aware template matching method is innovatively introduced, effectively suppressing erroneous matches through a quality metric mechanism, thereby significantly improving matching accuracy and localization stability while maintaining high robustness. Attached Figure Description

[0016] The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Throughout the drawings, the same reference numerals denote the same parts. Obviously, the drawings described below are merely some embodiments of the present invention, and those skilled in the art can obtain other drawings based on these drawings without any creative effort.

[0017] Figure 1 This is a flowchart illustrating a game icon positioning method based on feature vectors provided in an embodiment of the present invention.

[0018] Figure 2 This is a schematic diagram of a game icon positioning system based on feature vectors provided in an embodiment of the present invention. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solutions in the embodiments of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. It should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0020] The game icon localization method based on feature vectors provided by the present invention will be described in detail below with reference to the accompanying drawings, through specific embodiments and application scenarios.

[0021] Reference manual attached Figure 1 The diagram shows a flowchart of a game icon positioning method based on feature vectors provided by an embodiment of the present invention.

[0022] This invention provides a method for locating game icons based on feature vectors, which may include the following steps: S1: Get the game icon to be processed.

[0023] S2: Preprocess the game icon.

[0024] It should be noted that the preprocessed image is used to remove input images that are irrelevant to the game, so that the resolution can be uniformly corrected to a fixed resolution later.

[0025] In one possible implementation, S2 specifically includes sub-steps S201 to S204: S201: Scale the game icon to a preset uniform resolution.

[0026] It should be noted that the input image is scaled to a preset uniform resolution (such as 128×128) to eliminate the size difference of the icon under different acquisition conditions.

[0027] It should be noted that those skilled in the art can set the preset resolution size according to actual needs, and this invention does not limit that.

[0028] S202: Perform mean and variance normalization operations on the RGB channels of the game icon.

[0029] It should be noted that mean and variance normalization is performed on the RGB channels of the image to reduce the impact of lighting variations and device differences on the features.

[0030] S203: During the training phase, perform data augmentation operations on the game icons, including random rotation, translation, scaling, and blurring.

[0031] It should be noted that during the training phase, images are subjected to random rotation, translation, scaling, blurring, and other enhancement operations to improve the robustness of features to pose changes and noise interference.

[0032] S204: Generate multi-scale versions of game icons according to recognition needs.

[0033] It should be noted that multiple scale versions of the image are generated as needed to ensure that stable features can be extracted at different resolutions.

[0034] S3: Perform a sliding window operation on the preprocessed game icon to generate multiple candidate regions.

[0035] The sliding window operation refers to sliding on the input image with a preset step size and window size to capture image blocks that may contain the target icon, ensuring that no potential target locations are missed.

[0036] It should be noted that those skilled in the art can set the preset step size according to actual needs, and this invention does not limit this.

[0037] It should be noted that a sliding window operation is performed on the screenshot to be located to extract the area for subsequent feature comparison.

[0038] In this embodiment of the invention, the sliding window operation ensures the coverage of the detection. No matter where the target icon appears on the screen or what size it is, it can be included in the candidate area, effectively avoiding omissions and providing a complete candidate set for subsequent accurate positioning.

[0039] S4: Extract global and local feature vectors of candidate regions using the XFeat deep neural network.

[0040] Among them, XFeat deep neural network is a deep learning model specifically designed for efficient feature extraction; global feature vectors describe the overall outline and texture of the icon, while local feature vectors capture the key details and corners of the icon. The combination of the two can comprehensively and robustly represent the icon features.

[0041] It should be noted that the deep neural network XFeat is used as the feature extraction backbone to extract global and local feature vectors for each icon template and candidate screenshot area.

[0042] It should be noted that deep neural networks (such as the XFeat algorithm) are used to process the template icons to extract multi-scale global and local features. The extracted feature vectors are then normalized to improve the consistency of similarity calculations.

[0043] Specifically, the XFeat algorithm neural network is used to perform forward propagation on the template icon; through the forward operation of the neural network, the global features (describing the overall shape, color and texture of the icon) and local features (used to capture key local areas in the icon, such as edges, corners and symbols) of the template icon are extracted.

[0044] In this embodiment of the invention, the XFeat network is used to extract global and local features simultaneously, which can obtain robust feature representations that include both macroscopic contours and fine structures, thereby providing stronger resistance to local occlusion, deformation and background interference of icons.

[0045] S5: Standardize the global and local feature vectors to obtain standardized feature vectors.

[0046] Standardization usually refers to L2 normalization, which scales the feature vectors to a unit length. This eliminates differences in the magnitude of the feature vectors, making subsequent similarity calculations more fair and accurate.

[0047] In one possible implementation, S5 specifically includes sub-steps S501 to S504: S501: Normalize the extracted global feature vector: in, This represents the normalized global feature vector, ensuring a vector length of 1 to avoid similarity calculation errors caused by scale differences. Represents the global feature vector. express Norm, This represents the total dimension of the global feature vector. Indicates the first A global feature vector with multiple dimensions.

[0048] It should be noted that the global feature vector is a high-dimensional feature (such as a 512-dimensional vector) output by the network backbone, used for fast retrieval and coarse matching.

[0049] S502: Perform PCA dimensionality reduction on local feature vectors.

[0050] S503: Redundant feature filtering is performed on local feature vectors to remove local feature points with low confidence.

[0051] It should be noted that a set of feature points (such as a 128-dimensional vector) obtained through keypoint detection and local region encoding is used for subsequent fine matching and spatial location backtracking.

[0052] S504: Integrate the normalized global feature vector with the PCA-reduced local feature vector to obtain the standardized feature vector.

[0053] It should be noted that L2 normalization is performed on the global feature vector to facilitate retrieval using cosine similarity; PCA dimensionality reduction is performed on the local feature point descriptors to reduce storage and computational overhead while maintaining discriminative ability; and redundant features are filtered out to remove local points with low confidence, thereby improving matching accuracy.

[0054] In this embodiment of the invention, the normalization process eliminates the influence of the feature vector's magnitude, making the subsequent similarity calculation depend entirely on the differences in the feature direction. This significantly improves the fairness and accuracy of the similarity measurement, laying the foundation for high-quality matching.

[0055] S6: Perform quality similarity matching between the standardized feature vector and the template feature vector in the icon template library.

[0056] The icon template library is a database that stores all known icon feature vectors; quality similarity matching specifically refers to methods such as QATM (Quality Aware Template Matching), which not only calculates similarity but also evaluates the quality and uniqueness of the matching results, effectively suppressing false matches.

[0057] In one possible implementation, S6 specifically includes sub-steps S601 to S603: S601: Calculate the cosine similarity between the standardized feature vector and the template feature vector: in, Represents the standardized feature vector. Represents the template feature vector. The cosine similarity between two feature vectors is represented by the following expression: express Norm.

[0058] S602: Based on cosine similarity, calculate the quality assessment score using a quality metric function. in, This represents the quality assessment score, with the denominator being the normalized similarity result between the candidate region and all features in the template library. This represents an exponential function with base to natural numbers. This represents the parameter used to adjust the smoothness of the matching distribution. This indicates the total number of template feature vectors in the icon template library. Indicates the first icon template in the icon template library k Template feature vectors.

[0059] S603: Sort the quality assessment scores of all standardized feature vectors and select the standardized feature vector with the highest score that exceeds the preset quality assessment score threshold to obtain the matching result.

[0060] It should be noted that those skilled in the art can set the size of the preset threshold according to actual needs, and this invention does not limit this.

[0061] Specifically, if multiple candidate regions correspond to the same template icon category, then non-maximum suppression (NMS) is used to remove redundancy and obtain a unique localization result.

[0062] In this embodiment of the invention, a quality-aware mechanism is introduced, which goes beyond simple similarity calculation and can effectively distinguish between "true matches" and "accidental similarities", thereby accurately identifying the correct target among many similar icons and significantly reducing the false match rate.

[0063] Optionally, the construction process of the icon template library specifically includes: Collect game icon template data.

[0064] Specifically, icon collection involves pulling game image data and capturing corresponding asset icon elements such as game characters, equipment, and items. The goal is to comprehensively cover asset icons in different states to maximize coverage.

[0065] Each game icon template data collected is labeled with category and location information to obtain a standardized template dataset.

[0066] Specifically, for each collected icon, category and location information are labeled manually or semi-automatically to form a standardized template dataset.

[0067] Data augmentation operations are performed on the standardized template dataset, including scaling, geometric transformation, color perturbation, background blending, and transparency processing.

[0068] Specifically, scaling transformation: generates templates of different sizes by scaling to enhance cross-resolution adaptability; geometric transformation: includes rotation, translation, and mirroring to enhance the diversity of icon poses; color perturbation: adjusts brightness, saturation, and contrast to adapt to lighting differences in different game environments; background blending: overlays icons onto various random backgrounds to simulate interference factors in the actual game interface; transparency processing: retains the icon body and removes invalid background areas to reduce matching errors.

[0069] It should be noted that this enhances the robustness of the templates and the diversity of data during subsequent template matching. After the above processing, a high-quality, standardized template icon dataset is finally formed, providing reliable input for subsequent feature extraction and library construction.

[0070] The enhanced template dataset is normalized using the XFeat deep neural network to obtain standardized feature vectors.

[0071] The standardized feature vectors of all template datasets are used to form a feature set.

[0072] An approximate nearest neighbor index structure is used to store and retrieve feature sets.

[0073] Optionally, an approximate nearest neighbor index structure is used. The process of storing and retrieving the feature set specifically includes: It should be noted that all template feature vectors are stored in the index database, and the index is constructed using the efficient approximate nearest neighbor search method FAISS, which facilitates rapid template retrieval and subsequent expansion by adding new game templates.

[0074] Based on the cluster center set obtained in advance through clustering methods, calculate the Euclidean distance between the normalized feature vector and each cluster center, and select the clusters with the closest distance as candidate clusters: in, Indicates candidate clusters, This indicates the search for parameters that minimize the Euclidean distance. Represents the first cluster obtained through the clustering algorithm. Cluster center vectors, express Norm.

[0075] The Euclidean distances are sorted from highest to lowest, and the first preset number of clusters are selected as candidate clusters.

[0076] It should be noted that those skilled in the art can set the preset quantity according to actual needs, and this invention does not limit this.

[0077] It should be noted that by using clustering methods (such as K-means) to divide the standardized feature vectors, the space is divided into multiple cluster centers.

[0078] In the inverted index table corresponding to the candidate cluster, calculate the cosine similarity between the standardized feature vector and the template feature vector.

[0079] It should be noted that the standardized feature vectors are assigned to the corresponding clusters to form an inverted index table, which enables fast screening of the candidate set; Based on cosine similarity, a predetermined number of template feature vectors with the highest similarity are selected from the candidate clusters to obtain the feature set: in, Indicates the first result obtained from the search KThe set of the most similar template feature vectors.

[0080] It should be noted that those skilled in the art can set the preset quantity according to actual needs, and this invention does not limit this.

[0081] Clustering methods are used to divide the feature set into an inverted index table, enabling rapid screening of the candidate set.

[0082] It should be noted that the icon template library is designed to transform template data into a set of feature vectors that can be quickly retrieved, and to build an efficient index structure.

[0083] Furthermore, to further improve retrieval speed and matching accuracy, this invention introduces the following optimization strategies during the index construction process: Feature Hierarchy: Icons are pre-grouped according to category, and an index structure is built within each group to reduce global search overhead; Multi-Scale Indexing: Feature indexes are built separately for templates of different sizes to achieve multi-scale parallel retrieval; Dynamic Updates: Online feature insertion is supported when adding new icon templates without rebuilding the entire database; Redundancy Validation: Fine calculations of cosine similarity or Euclidean distance are introduced into the initial search results to ensure the accuracy of the final match. Through the above steps, the constructed feature library has high robustness, high scalability, and high retrieval efficiency, and can serve as the core support for subsequent matching and localization modules.

[0084] S7: Perform post-processing on the matching results, including nonmaximum suppression, multi-scale fusion, and confidence filtering, to obtain optimized matching results.

[0085] Among them, non-maximum suppression is used to eliminate overlapping redundant detection boxes; multi-scale fusion integrates the recognition results under different scaling ratios; and confidence filtering filters out low-quality matches based on the score threshold.

[0086] Specifically, fusing multi-scale matching results can improve the recognition rate of small icons; applying non-maximum suppression to the matching results can eliminate duplicate detections; and optimizing the matching results according to game rules or category constraints can improve stability. Finally, an optimal matching category result is returned.

[0087] In one possible implementation, S7 specifically includes sub-steps S701 to S703: S701: Perform nonmaximum suppression operation on the matching result.

[0088] In one possible implementation, S701 specifically includes sub-steps S7011 to S7013: S7011: For different candidate detection boxes corresponding to the quality similarity matching results, calculate the intersection-union ratio (IUU) between different candidate detection boxes: in, Indicates intersection, union, and ratio. Indicates the first One candidate detection box, Indicates the first One candidate detection box, This represents the area of ​​the intersection and union of candidate detection boxes. This represents the intersection region of two candidate detection boxes. This represents the union region of two candidate detection boxes.

[0089] It should be noted that Non-maximum Suppression (NMS) calculates the overlap of multiple detection results of the same icon category in adjacent regions. If the Intersection over Union (IoU) of two candidate results is greater than the preset IoU threshold, the result with the highest confidence is retained and redundant detection boxes are removed, thereby avoiding duplicate output.

[0090] S7012: Sort all candidate detection boxes by quality assessment score.

[0091] S7013: Select the candidate detection box with the highest score after sorting as the reference box, calculate the intersection-union ratio (IUR) of the remaining candidate detection boxes with the reference box, and remove candidate detection boxes with IUR greater than the preset IUR threshold.

[0092] It should be noted that those skilled in the art can set the preset crossover ratio according to actual needs, and this invention does not limit it.

[0093] S702: Perform a multi-scale fusion operation on the candidate detection boxes after non-maximum suppression, and perform weighted voting on the candidate detection boxes from different scales to generate the final localization result.

[0094] It should be noted that, since candidate regions may come from sliding windows of different scales, this invention fuses the detection results at multiple scales; for multi-scale detection at the same location, a weighted voting method is used to generate the final result, ensuring robustness at different resolutions.

[0095] S703: Perform a confidence filtering operation on the final location results, filtering out similarity matching results with low confidence based on a preset confidence threshold.

[0096] It should be noted that, to avoid low-quality matching results interfering with the final output, this invention sets a confidence threshold in post-processing; only the Top K results with a confidence level greater than the threshold are output, ensuring that the final localization result achieves a balance between accuracy and recall. This invention can effectively reduce redundant detection, improve the stability of cross-scale recognition, and eliminate low-confidence results, thereby significantly improving the overall accuracy and robustness of game icon matching and localization.

[0097] In this embodiment of the invention, through multi-level post-processing, this step can effectively eliminate duplicate detections, integrate the advantages of multiple scales, and filter low-confidence results, ultimately outputting an optimized result that is accurate in location, unique in category, and highly reliable, greatly improving the accuracy of positioning and the stability of output.

[0098] S8: Based on the optimized matching results, output the location and category information of the game icon to complete the positioning of the game icon.

[0099] In this embodiment of the invention, a clear and usable final output (what it is and where it is) is provided, which directly transforms the recognition results of this method into precise instructions required by downstream applications (such as automatic operation and asset statistics), realizing a closed loop from image understanding to practical application.

[0100] Optionally, after S8, it also includes: S9: When a match fails, the icon template library is dynamically updated based on the standardized feature vector.

[0101] It should be noted that the template library can be dynamically updated. When there is no relevant template in the template library and the matching fails, a new label can be added to the image feature and added to the template library for subsequent matching.

[0102] The dynamic template library update technology is implemented as follows: Matching failure determination: After calculating the feature similarity between the target region and the preset templates in the template library, if the similarity score is lower than the preset threshold, it is determined as a matching failure.

[0103] It should be noted that those skilled in the art can set the size of the preset threshold according to actual needs, and this invention does not limit this.

[0104] Candidate region extraction: For regions deemed unmatchable, feature enhancement and noise reduction are performed to obtain feature vectors for candidate regions. These candidate regions, meeting the criteria of clarity, completeness, and feature saliency, are selected as the source for template updates.

[0105] Template library update strategy: If the category corresponding to a candidate region does not exist in the template library, the region's features and its corresponding image are directly added to the template library. If a template for that category already exists in the template library, but the similarity remains below the threshold, the original template is replaced or updated using newly extracted candidate region features through weighted fusion. To avoid template library homogenization, the system can store multiple representative templates of the same category, using a queue or sliding window mechanism to maintain template diversity.

[0106] It should be noted that dynamically updating the template library is used to improve the system's adaptability and robustness.

[0107] The beneficial effects of the technical solutions provided by the embodiments of the present invention include at least the following: In this embodiment of the invention, robust global and local features are extracted using an XFeat deep neural network, and multi-dimensional data augmentation is combined to effectively overcome the sensitivity issues caused by icon scale, rotation, and background changes. To address the template library scalability issue, an efficient vector index based on FAISS is constructed with a dynamic update mechanism, enabling rapid expansion of the template library without retraining. Furthermore, a QATM quality-aware template matching method is innovatively introduced, effectively suppressing erroneous matches through a quality metric mechanism, thereby significantly improving matching accuracy and localization stability while maintaining high robustness.

[0108] The game icon localization method based on feature vectors provided in this application can be executed by a game icon localization device based on feature vectors. This application uses the example of a game icon localization device based on feature vectors executing the game icon localization method based on feature vectors to illustrate the game icon localization device based on feature vectors provided in this application.

[0109] Reference manual attached Figure 2 The diagram shows a schematic representation of a game icon positioning device based on feature vectors provided in an embodiment of the present invention.

[0110] This invention provides a game icon positioning system 20 based on feature vectors, including: a processor 201 and a memory 202; The memory 202 stores programs or instructions that can run on the processor 201. When the program or instructions are executed by the processor 201, they implement the steps of the above-described game icon positioning method based on feature vectors and achieve the same technical effect. To avoid repetition, the present invention will not elaborate further.

[0111] It should be understood that the processor 201 in this embodiment of the invention may be a central processing unit (CPU), or it may be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0112] It should also be understood that the memory 202 in the embodiments of the present invention can be volatile memory or non-volatile memory, or may include both volatile and non-volatile memory. The non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. The volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of random access memory are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct memory bus RAM (DR RAM).

[0113] The above embodiments can be implemented, in whole or in part, by software, hardware (such as circuits), firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of the present invention are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0114] It should be understood that, in various embodiments of the present invention, the order of the above-mentioned process numbers does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0115] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0116] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the devices, apparatuses, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0117] In the several embodiments provided by this invention, it should be understood that the disclosed devices, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another device, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0118] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0119] In addition, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0120] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0121] This invention provides a readable storage medium that stores a program or instructions on the storage medium. When the program or instructions are executed by a processor, they implement the steps of the above-described game icon positioning method based on feature vectors and achieve the same technical effect. To avoid repetition, this invention will not elaborate further.

[0122] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the embodiments of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the protection scope of the present invention.

Claims

1. A feature vector based game icon positioning method, characterized by, The method comprises the following steps: S1: obtaining a game icon to be processed; S2: preprocessing the game icon; S3: performing a sliding window operation on the preprocessed game icon to generate a plurality of candidate regions; S4: extracting global feature vectors and local feature vectors of the candidate regions through an XFeat deep neural network; S5: normalizing the global feature vectors and the local feature vectors to obtain standardized feature vectors; S6: performing quality similarity matching on the standardized feature vectors and template feature vectors in an icon template library; S7: performing post-processing including non-maximum suppression, multi-scale fusion and confidence filtering on the matching results to obtain optimized matching results; S8: outputting position information and category information of the game icon according to the optimized matching results to complete positioning of the game icon.

2. The feature vector based game icon positioning method of claim 1, wherein, The S2 specifically comprises: S201: scaling the game icon to a preset uniform resolution; S202: performing mean and variance normalization operations on the RGB channels of the game icon; S203: performing data enhancement operations on the game icon in the training stage, wherein the data enhancement operations include random rotation, translation, scaling and blurring; S204: generating multi-scale versions of the game icon according to the identification needs.

3. The feature vector based game icon positioning method of claim 1, wherein, The S5 specifically comprises: S501: normalizing the extracted global feature vectors; wherein, denotes the normalized global feature vector, denotes the global feature vector, denotes norm, denotes the total dimension of the global feature vector, denotes the global feature vector of the th dimension; S502: performing PCA dimensionality reduction processing on the local feature vectors; S503: performing redundant feature screening on the local feature vectors to eliminate low-confidence local feature points; S504: integrating the normalized global feature vectors and the PCA dimensionality reduced local feature vectors to obtain the standardized feature vectors.

4. The feature vector based game icon positioning method of claim 1, wherein, The S6 specifically comprises: S601: calculating the cosine similarity between the standardized feature vectors and the template feature vectors; wherein, denotes a normalized feature vector, denotes a template feature vector, denotes a cosine similarity between two feature vectors, denotes norm; S602: calculating the quality evaluation score through a quality measurement function based on the cosine similarity; wherein, denotes the quality assessment score, denotes the exponential function with base natural number, denotes a parameter for adjusting the smoothness of the matching distribution, denotes the total number of template feature vectors in the icon template library, denotes the i-th template feature vector in the icon template library, k denotes the i-th template feature vector in the icon template library. S603: sorting the quality evaluation scores of all standardized feature vectors, and selecting the standardized feature vectors with the highest scores and exceeding a preset quality evaluation score threshold to obtain the matching results.

5. The feature vector based game icon positioning method of claim 1, wherein, The construction process of the icon template library in S6 specifically comprises: collecting game icon template data; annotating the collected game icon template data with category and position information to obtain a standardized template data set; performing data enhancement operations on the standardized template data set, wherein the data enhancement operations include scale transformation, geometric transformation, color disturbance, background fusion and transparent processing; performing normalization processing on the enhanced template data set through an XFeat deep neural network to obtain the standardized feature vectors; constructing a feature set from the standardized feature vectors of all template data sets; storing and retrieving the feature set by using an approximate nearest neighbor index structure; dividing the feature set by using a clustering method to form an inverted index table and realize fast screening of the candidate set.

6. The feature vector based game icon positioning method of claim 5, wherein, The process of storing and retrieving the feature set by using an approximate nearest neighbor index structure specifically comprises: According to a cluster center set obtained in advance by a clustering method, a Euclidean distance between a standardized feature vector and each cluster center is calculated: wherein, denotes a candidate cluster, denotes finding the parameters that minimize the Euclidean distance, denotes the k-th cluster center vector obtained by the clustering algorithm, denotes the k-th cluster center vector obtained by the clustering algorithm, denotes norm; In a descending order, each of the Euclidean distances is sorted, and a preset number of clusters are selected as candidate clusters; In an inverted index table corresponding to the candidate clusters, a cosine similarity between the standardized feature vector and the template feature vector is calculated; According to the cosine similarity, a preset number of template feature vectors with the highest similarity are selected from the candidate clusters, and the feature set is obtained: wherein, denotes the retrieved top K most similar template feature vector sets.

7. The feature vector based game icon positioning method of claim 1, wherein, The S7 specifically includes: S701: performing a non-maximum suppression operation on the matching result; S702: performing a multi-scale fusion operation on the candidate detection frame after the non-maximum suppression, and generating a final positioning result by weighted voting of the candidate detection frames from different scales; S703: performing a confidence filtering operation on the final positioning result, and filtering the similarity matching result with low confidence according to a preset confidence threshold.

8. The feature vector based game icon positioning method of claim 7, wherein, The S701 specifically includes: S7011: calculating an intersection-over-union between different candidate detection frames corresponding to the quality similarity matching result; wherein, denotes the intersection over union, denotes the i-th candidate bounding box, denotes the i-th candidate bounding box, denotes the i-th candidate bounding box, denotes the i-th candidate bounding box, denotes the intersection over union between two candidate bounding boxes, denotes the intersection area of two candidate bounding boxes, denotes the union area of two candidate bounding boxes; S7012: sorting all the candidate detection frames by the quality evaluation score; S7013: selecting a candidate detection frame with the highest score after sorting as a reference frame, calculating an intersection-over-union between the reference frame and the remaining candidate detection frames, and removing the candidate detection frame with an intersection-over-union greater than a preset intersection-over-union threshold.

9. The feature vector based game icon positioning method of claim 1, wherein, The S8 further includes: S9: when the matching fails, dynamically updating the icon template library according to the standardized feature vector.

10. A feature vector based game icon positioning system, comprising: It includes: a processor and a memory; The memory stores programs or instructions executable on the processor, and the programs or instructions are executed by the processor to implement the steps of the feature vector-based game icon positioning method according to any one of claims 1 to 9.

Images