Fast image recognition method capable of saving computing power resources

Abstract

The invention discloses a fast image recognition method capable of saving computing power resources. The fast image recognition method comprises the following steps of 1, performing wavelet decomposition on a gray histogram of an original image; step 2, filtering the signal through a digital low-pass filter and a high-pass filter, namely LoD and HiD; step 3, even number point sampling is carried out on a filtering result, so that smooth approximation and detail information of a next level are obtained; and 4, calculating a threshold value by using an improved maximum between-class variance method. The invention belongs to the technical field of image recognition, and effectively solves the problems of high calculation complexity, large resource consumption and insufficient real-time performance in the existing rapid image recognition method.

Description

Technical Field

[0001] The present invention belongs to the technical field of image recognition, and specifically refers to a fast image recognition method that saves computing resources. Background Art

[0002] With the rapid development of artificial intelligence technology, the demand for computing resources in image recognition tasks has increased dramatically. Traditional image segmentation methods, such as grayscale histogram-based threshold segmentation, require traversing all grayscale levels to search for the optimal threshold, which is computationally complex and time-consuming, especially at high grayscale levels, which will significantly extend the operation time. Although there are improved algorithms, such as the maximum inter-class variance method, which improves segmentation accuracy, the amount of calculation still increases linearly with the number of grayscale levels, making it difficult to meet real-time requirements.

[0003] In addition, existing methods still have shortcomings in multi-resolution analysis and fast segmentation, and the computing power consumption is particularly prominent, which cannot meet the requirements of real-time and high efficiency.

[0004] In recent years, wavelet analysis has been introduced into the field of image processing due to its multi-resolution characteristics, but existing methods focus more on reconstruction accuracy and do not fully explore its potential in reducing computational complexity. Therefore, there is an urgent need for an image recognition method that can effectively narrow the threshold search range, reduce computing power consumption, and quickly segment the target and background.

[0005] The implementation scheme most similar to the present invention is based on the maximum inter-class variance method, namely the traditional image segmentation technology of the Otsu method. This method determines the optimal threshold by maximizing the inter-class variance and is widely used in image segmentation. This method has the following main disadvantages: 1. High computational complexity: The traditional maximum inter-class variance method needs to calculate the inter-class variance for each possible threshold. Especially at high gray levels, such as L=256, the amount of calculation increases significantly, resulting in longer processing time. 2. High resource consumption: This method usually needs to process high-resolution or high-grayscale images, which will increase the consumption of computing resources and limit its application in resource-constrained environments; 3. Lack of real-time performance: In industrial inspection, on-site identification and other scenarios, image recognition requires real-time feedback. However, due to the high computational complexity, traditional methods often cannot achieve fast response, affecting the overall efficiency. Summary of the invention

[0006] The technical problem to be solved by the present invention is that the existing fast image recognition methods have high computational complexity, large resource consumption, and insufficient real-time performance.

[0007] The technical solution adopted by the present invention is as follows: A fast image recognition method for saving computing resources proposed by the present invention comprises the following steps: Step 1: Perform wavelet decomposition on the grayscale histogram of the original image; Step 2: Filter the signal through digital low-pass filter and high-pass filter, namely Lo_D and Hi_D; Step 3: Sample the filtering result at even points to obtain the next level of smooth approximation and detail information; Step 4: Calculate the threshold using the improved maximum inter-class variance method.

[0008] Furthermore, the low-pass filter coefficient in step 2 is [0.7071, 0.7071].

[0009] Furthermore, in step 4, the segmentation threshold is calculated using the improved maximum inter-class variance method on the low-frequency component of the original image after wavelet decomposition. The specific calculation formula is as follows: ; .

[0010] Furthermore, in the calculation formula, At and Bt are the total number of pixels and the sum of gray levels of region C0, An and Bn are the total number of pixels and the sum of gray levels of the entire image, and th is the optimal segmentation threshold calculated after wavelet decomposition of the image.

[0011] Furthermore, according to the scaling characteristics of wavelet transform, the calculated threshold is interpolated and reconstructed to obtain the segmentation threshold of the original image.

[0012] The beneficial effects achieved by the present invention using the above structure are as follows: 1. Improved computing efficiency: Through wavelet decomposition and multi-resolution analysis, the complexity of the grayscale histogram is significantly reduced, the calculation steps are reduced, and the image processing speed is improved.

[0013] 2. Reduced resource consumption: It avoids complex probability calculations and reconstruction of detailed information, reduces dependence on computing resources, and is particularly suitable for embedded or resource-constrained systems.

[0014] 3. Enhanced real-time performance: By optimizing the algorithm and reducing the number of calculation steps, the real-time performance of the method is improved, meeting the needs of intelligent driving, industrial detection and other fields for real-time feedback.

[0015] 4. High precision: The scale expansion characteristics of wavelet transform and the improved inter-class variance calculation method ensure the high precision of the segmentation results and improve the accuracy of overall recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 Schematic diagram of wavelet decomposition of image grayscale histogram in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0018] Embodiment 1: A fast image recognition method for saving computing resources proposed by the present invention comprises the following steps: Step 1: Wavelet decomposition of image grayscale histogram; Step 2: Filter the signal through a digital low-pass filter and a high-pass filter; Step 3: Sample the filtering result at even points to obtain the next level of smooth approximation and detail information; Step 4: Calculate the threshold using the improved maximum inter-class variance method.

[0019] The low-pass filter coefficient of step 2 is [0.7071, 0.7071].

[0020] In step 4, the segmentation threshold is calculated using the improved maximum inter-class variance method on the low-frequency component of the original image after wavelet decomposition. The calculation of the probability of each pixel value is avoided by simplifying the formula, which significantly improves the calculation speed. The specific calculation formula is as follows: ; ; In the calculation formula, At and Bt are the total number of pixels and the sum of gray levels in area C0 respectively, An and Bn are the total number of pixels and the sum of gray levels in the whole image, and th is the optimal segmentation threshold calculated after wavelet decomposition of the image; for a known image, the value of An is fixed, so when calculating the segmentation threshold, there is no need to calculate the probability of each pixel value, but only two statistics At and Bt need to be calculated.

[0021] According to the scaling characteristics of wavelet transform, the calculated threshold is interpolated and reconstructed to obtain the segmentation threshold of the original image; In specific use, for example, the original image is subjected to r wavelet transforms, and after the threshold t is calculated, according to the scale scalability of the wavelet transform, the threshold of the original image should be T = t × 2 r .

[0022] This method ensures high accuracy of the segmentation results while avoiding the reconstruction of detail information and significantly reducing the computational burden.

[0023] Through the above steps, the present invention can significantly reduce the consumption of computing resources while ensuring recognition accuracy, and is particularly suitable for embedded or resource-constrained image processing systems.

[0024] In summary, when the embodiment of the present invention is used, the step of calculating the image segmentation threshold is to first perform wavelet decomposition, use the improved maximum inter-class variance method to find the threshold, and then reconstruct the threshold segmentation; In the process of wavelet decomposition of image histogram, only low-frequency components are calculated, and high-frequency components are not calculated, which reduces the consumption of computing resources; The improved segmentation threshold calculation algorithm no longer needs to calculate the probability of each pixel value, which significantly reduces the amount of calculation.

[0025] The above is the entire usage process of a fast image recognition method that saves computing resources.

[0026] The present invention and its implementation methods are described above, and such description is not restrictive, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design structures and embodiments similar to the technical solution without creativity, which should all fall within the protection scope of the present invention.

Claims

1. A fast image recognition method that saves computing resources, characterized in that: The following steps are involved: Step 1: Perform wavelet decomposition on the grayscale histogram of the original image; Step 2: Filter the signal through digital low-pass filter and high-pass filter, namely Lo_D and Hi_D; Step 3: Sample the filtering result at even points to obtain the next level of smooth approximation and detail information; Step 4: Calculate the threshold using the improved maximum inter-class variance method.

2. The method for rapid image recognition that saves computing resources according to claim 1, characterized in that: The low-pass filter coefficient of step 2 is [0.7071, 0.7071].

3. The method for rapid image recognition that saves computing resources according to claim 1, characterized in that: In step 4, the segmentation threshold is calculated using the improved maximum inter-class variance method on the low-frequency component of the original image after wavelet decomposition. The specific calculation formula is as follows: ； 。 4. The method for rapid image recognition that saves computing resources according to claim 3 is characterized in that: In the calculation formula, At and Bt are the total number of pixels and the sum of gray levels in area C0, An and Bn are the total number of pixels and the sum of gray levels in the whole image, and th is the optimal segmentation threshold calculated after wavelet decomposition of the image.

5. The method for fast image recognition that saves computing resources according to any one of claims 1 to 4, characterized in that: According to the scaling characteristics of wavelet transform, the calculated threshold is interpolated and reconstructed to obtain the segmentation threshold of the original image.

Images