Multi-algorithm fusion intelligent image annotation system and method based on real-time image enhancement

Abstract

The invention provides a multi-algorithm fusion intelligent image annotation system and method based on real-time image enhancement. Comprising a real-time image enhancement module, a multi-algorithm fusion module, an intelligent labeling module and a synchronous processing module. Wherein the real-time image enhancement module is used for optimizing an input low-quality image in real time; the multi-algorithm fusion module is used for uniformly integrating and dynamically calling multiple target detection and tracking algorithms; the intelligent labeling module is used for acquiring a target detection or tracking algorithm output result of the multi-algorithm fusion module in real time and completing multi-target positioning and labeling; the synchronous processing module comprises a multi-thread task scheduling unit and a data stream cache unit; the multi-thread task scheduling unit is used for coordinating parallel execution of an image enhancement task, an algorithm calling task, a labeling task and an interface updating task; and the data stream caching unit is used for caching data of each module in real time.

Description

Technical Field

[0001] This invention belongs to the field of computer vision and artificial intelligence, and specifically relates to a multi-algorithm fusion intelligent image annotation system and method based on real-time image enhancement. Background Technology

[0002] With the rapid development and widespread application of artificial intelligence technology, the demand for high-quality, large-scale data annotation in the field of computer vision is becoming increasingly prominent. Especially in applications such as object detection and object recognition, high-precision image annotation data is the foundation and key to model training and performance improvement. However, currently widely used mainstream image annotation tools, such as LabelImg, CVAT, and VIA, generally have significant technical bottlenecks and shortcomings, mainly reflected in the following aspects: First, these traditional annotation tools are relatively limited in their algorithmic functionality, typically supporting only one specific type or purpose of algorithm, and cannot adapt to diverse and complex annotation tasks simultaneously. Real-world scenarios often require the simultaneous identification and annotation of multiple different target types, and single-algorithm tools struggle to flexibly meet this need, significantly reducing data processing efficiency and dataset diversity.

[0003] Secondly, current mainstream annotation tools lack effective real-time image quality optimization methods. This is especially true when dealing with low-quality images, such as those in low-light conditions, blurred images, or those affected by noise, resulting in extremely poor annotation quality and severely impacting annotation accuracy and reliability. Furthermore, existing annotation tools typically separate image preprocessing and image annotation into independent processes, failing to enable real-time interaction and feedback adjustments. Users struggle to optimize image quality during the annotation process, leading to low efficiency and a cumbersome workflow.

[0004] Furthermore, most existing tools lack a unified algorithm integration and management framework, making algorithm invocation and switching complex, time-consuming, and labor-intensive. Significant differences exist between the interfaces of different algorithms, and the lack of a unified interface standard creates substantial technical barriers for users when switching algorithms to meet different annotation needs, reducing the usability and efficiency of the tools.

[0005] In addition, traditional tools have a low overall level of intelligence and lack intelligent auxiliary annotation and automatic annotation mechanisms, which makes the annotation process too dependent on manual intervention, making it difficult to improve efficiency and meet the urgent needs of large-scale data annotation work.

[0006] Therefore, there is an urgent need to develop a more intelligent, efficient, and automated image annotation system and method that innovatively integrates real-time image enhancement technology with a multi-algorithm dynamic calling mechanism. This would improve image quality in real time and flexibly switch the most suitable intelligent algorithm according to the annotation scenario, thereby achieving intelligent annotation assistance. This would effectively overcome the above-mentioned technical defects and significantly improve the efficiency and accuracy of image annotation work. Summary of the Invention

[0007] The main objective of this invention is to overcome the problems of limited functionality, separation of image enhancement and annotation processes, complex algorithm calling methods, and insufficient system intelligence in the existing technology, and to provide a multi-algorithm fusion intelligent image annotation system and method based on real-time image enhancement.

[0008] To achieve the above objectives, the technical solution of the present invention is as follows.

[0009] A multi-algorithm fusion intelligent image annotation system based on real-time image enhancement includes: a real-time image enhancement module, a multi-algorithm fusion module, an intelligent annotation module, and a synchronous processing module; wherein: Real-time image enhancement module: used to optimize low-quality input images in real time; specifically, it uses an adaptive color level algorithm to adaptively adjust the black and white thresholds by statistically analyzing the percentiles of the image brightness distribution, and uses a multi-threaded image processing engine to divide the image into multiple sub-image blocks for parallel processing and then merges and outputs them to the multi-algorithm fusion module. Multi-algorithm fusion module: used to integrate and dynamically call multiple target detection and tracking algorithms, including a unified interface and algorithm adapters; the unified interface is used to uniformly manage and call local or remote target detection and tracking algorithms, and the algorithm adapter dynamically loads different algorithm components through a factory pattern and automatically switches between different types of target detection and tracking algorithms in real time; Intelligent annotation module: used to acquire the target detection or tracking algorithm output results of the multi-algorithm fusion module in real time and complete multi-target localization and annotation; Synchronization processing module: includes a multi-threaded task scheduling unit and a data stream caching unit; The multi-threaded task scheduling unit is used to coordinate the parallel execution of image enhancement tasks, algorithm calling tasks, annotation tasks, and interface update tasks; the data stream caching unit is used to cache data from each module in real time.

[0010] The beneficial effects of this invention are: 1. This invention integrates real-time image quality optimization technology with multiple target detection and tracking algorithms, which can not only effectively improve the annotation accuracy of low-quality images such as low light, noise, and blur, but also support real-time dynamic switching and efficient collaboration of multiple intelligent algorithms, greatly improving the flexibility and accuracy of annotation work. 2. This invention uses real-time image enhancement technology to dynamically optimize low-quality input images, giving them higher visual quality and clarity, thereby effectively reducing the annotation difficulties caused by image quality problems such as low light, blur, and noise. 3. This invention constructs a unified algorithm calling framework, integrates multiple mainstream target detection and tracking algorithms, and selects the optimal algorithm by labeling the scene, which greatly improves the accuracy, stability and efficiency of image labeling; 4. This invention further realizes real-time synchronous interaction between image enhancement, algorithm calling and annotation data processing, which significantly improves the overall efficiency of image annotation work and meets the urgent need for high-precision, large-scale dataset production. Attached Figure Description

[0011] Figure 1 This is a block diagram of the multi-algorithm fusion intelligent image annotation system based on real-time image enhancement according to the present invention; Figure 2 This is a flowchart of the real-time image enhancement module in this invention; Figure 3 This is a flowchart of the multi-algorithm fusion module in this invention. Detailed Implementation

[0012] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0013] like Figure 1 As shown in this embodiment, a multi-algorithm fusion intelligent image annotation system based on real-time image enhancement specifically includes: a real-time image enhancement module, a multi-algorithm fusion module, an intelligent annotation module, and a synchronous processing module; wherein: Real-time image enhancement module: used to optimize low-quality input images in real time; specifically, it uses an adaptive color level algorithm to adaptively adjust the black and white thresholds by statistically analyzing the percentiles of the image brightness distribution, and uses a multi-threaded image processing engine to divide the image into multiple sub-image blocks for parallel processing and then merges and outputs them to the multi-algorithm fusion module. In this embodiment, the adaptive color level algorithm adaptively adjusts the black and white thresholds by statistically analyzing the percentiles of the image brightness distribution. The specific steps are as follows: ① Convert the input image to grayscale; ② Calculate the brightness distribution using statistical grayscale histograms; ③ Calculate the 2nd percentile of the image brightness as the black threshold and the 98th percentile as the white threshold; ④ Adaptively adjust the dynamic range of image brightness based on the calculated threshold; this method can significantly improve the overall image clarity and visual effect, and eliminate the annotation difficulties caused by images that are too dark or too bright.

[0014] In this embodiment, to achieve real-time processing performance, a multi-threaded image processing engine is used to divide the image into multiple sub-image blocks for parallel processing. The specific steps are as follows: ① Divide the original image into multiple sub-image blocks and distribute them to multiple threads for processing in parallel; ② Each thread uses an adaptive color level algorithm to calculate local optimization parameters and perform gamma correction. In this embodiment, a lookup table (LUT) technique is used for gamma correction, which can further improve the local contrast and visual quality of image blocks. ③ After processing, merge the sub-image blocks to form an optimized complete image.

[0015] Multi-algorithm fusion module: used to integrate and dynamically call multiple target detection and tracking algorithms, including a unified interface and algorithm adapters; the unified interface is used to uniformly manage and call local or remote target detection and tracking algorithms, and the algorithm adapter dynamically loads different algorithm components through the factory pattern and automatically switches between different types of target detection and tracking algorithms in real time. This method can realize hot-swapping of algorithms and significantly reduce the cost and complexity of algorithm replacement.

[0016] In this embodiment, the unified interface is a REST API interface.

[0017] Intelligent annotation module: used to acquire the target detection or tracking algorithm output results of the multi-algorithm fusion module in real time and complete multi-target localization and annotation; Synchronization processing module: includes a multi-threaded task scheduling unit and a data stream caching unit; The multi-threaded task scheduling unit is used to coordinate the parallel execution of image enhancement tasks, algorithm calling tasks, annotation tasks, and interface update tasks. This approach ensures the overall efficient operation of the system and the absence of blocking between tasks, thereby comprehensively improving real-time processing performance and user interaction experience.

[0018] The data stream caching unit is used to cache data from each module in real time, which can reduce interaction latency.

[0019] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A multi-algorithm fusion intelligent image annotation system based on real-time image enhancement, characterized in that, include: The module includes a real-time image enhancement module, a multi-algorithm fusion module, an intelligent annotation module, and a synchronous processing module; among which: Real-time image enhancement module: used to optimize low-quality input images in real time; specifically, it uses an adaptive color level algorithm to adaptively adjust the black and white thresholds by statistically analyzing the percentiles of the image brightness distribution, and uses a multi-threaded image processing engine to divide the image into multiple sub-image blocks for parallel processing and then merges and outputs them to the multi-algorithm fusion module. Multi-algorithm fusion module: used to integrate and dynamically call multiple target detection and tracking algorithms, including a unified interface and algorithm adapters; the unified interface is used to uniformly manage and call local or remote target detection and tracking algorithms, and the algorithm adapter dynamically loads different algorithm components through a factory pattern and automatically switches between different types of target detection and tracking algorithms in real time; Intelligent annotation module: used to acquire the target detection or tracking algorithm output results of the multi-algorithm fusion module in real time and complete multi-target localization and annotation; Synchronization processing module: includes a multi-threaded task scheduling unit and a data stream caching unit; The multi-threaded task scheduling unit is used to coordinate the parallel execution of image enhancement tasks, algorithm calling tasks, annotation tasks, and interface update tasks; the data stream caching unit is used to cache data from each module in real time.

2. The intelligent image annotation system based on real-time image enhancement and multi-algorithm fusion as described in claim 1, characterized in that, The adaptive color level algorithm is used to adaptively adjust the black and white thresholds by statistically analyzing the percentiles of the image brightness distribution. The specific steps are as follows: ① Convert the input image to grayscale; ② Calculate the brightness distribution using statistical grayscale histograms; ③ Calculate the 2nd percentile of the image brightness as the black threshold and the 98th percentile as the white threshold; ④ Adaptively adjust the dynamic range of image brightness based on the calculated threshold.

3. A multi-algorithm fusion intelligent image annotation system based on real-time image enhancement as described in claim 1 or 2, characterized in that, A multi-threaded image processing engine is used to divide the image into multiple sub-image blocks for parallel processing. The specific steps are as follows: ① Divide the original image into multiple sub-image blocks and distribute them to multiple threads for processing in parallel; ② In each thread, the adaptive color level algorithm is used to calculate the local optimization parameters and perform gamma correction; ③ After processing, the sub-image blocks are merged to form the optimized complete image.

4. The intelligent image annotation system based on real-time image enhancement and multi-algorithm fusion as described in claim 3, characterized in that, Gamma correction is performed using a lookup table technique.

5. A multi-algorithm fusion intelligent image annotation system based on real-time image enhancement as described in claim 1, 2, or 3, characterized in that, The unified interface is the REST API interface.

Images