Appearance and trademark image retrieval similarity judgment method based on electric signal analysis
By fusing visual and electrical signal features and dynamically adjusting the weights, the accuracy and precision issues of traditional methods in appearance and trademark image similarity retrieval are solved, achieving more efficient similarity determination.
Patent Information
- Application Number
- CN202511135272.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-08-14
- Publication Date
- 2025-11-14
AI Technical Summary
Traditional methods for searching and determining the similarity of appearance and trademark images are not accurate enough when faced with complex images, struggle to capture subtle differences, and fail to fully utilize electrical signal features, resulting in bias and low accuracy in similarity judgment.
By fusing visual and electrical signal features, a deep learning model is used to extract visual and electrical signal features. The feature weights are dynamically adjusted based on the complexity of the image content and the stability of the electrical signal. An adaptive weight adjustment mechanism is then used to determine similarity.
It improves the accuracy and reliability of appearance and trademark image similarity determination, can more comprehensively and flexibly reflect image characteristics, and enhances the accuracy of determination.
Smart Images
Figure CN120953640A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of image retrieval technology, specifically to a method for determining the similarity of appearance and trademark images based on electrical signal analysis. Background Technology
[0002] In today's business activities and intellectual property protection fields, the importance of trademark registration, infringement determination, and product design novelty assessment is becoming increasingly prominent. In these key business scenarios, similarity retrieval and determination of appearance and trademark images are core links. The results directly affect key issues such as whether the trademark can be successfully registered, whether infringement can be accurately identified, and whether the product design is innovative. Therefore, developing efficient and accurate appearance and trademark image similarity retrieval and determination technologies is of vital importance for maintaining market order and promoting innovative development.
[0003] Traditional methods for retrieving and determining the similarity of appearance and trademark images primarily rely on visual features. However, this approach exhibits several limitations in practical applications. Firstly, its accuracy is severely insufficient for complex images. When faced with images rich in detail, complex textures, or unique color combinations, traditional visual feature extraction methods often struggle to comprehensively and accurately capture key information, leading to biased similarity judgments. Secondly, traditional methods fail to capture subtle differences within images. In appearance and trademark images, subtle shape variations, color gradations, or texture differences can significantly impact similarity determination, but traditional visual feature analysis methods have low sensitivity to these subtle features, easily overlooking crucial information. Furthermore, existing technologies do not fully utilize the electrical signal features in images. Image pixels not only possess visual attributes but also contain rich electrical properties, which can reflect image information from another dimension. Simultaneously, traditional methods employ fixed or manually set feature weight allocation, failing to fully consider the dynamic changes in image content and electrical signal features. This severely limits the comprehensiveness and accuracy of similarity determination, failing to meet the demands for high-precision and high-reliability similarity retrieval in practical applications. Summary of the Invention
[0004] The purpose of this invention is to overcome the shortcomings of existing technologies and provide a method for determining the similarity of appearance and trademark images based on electrical signal analysis. This method can fully utilize the advantages of both visual and electrical signal features by fusing visual features and electrical signal features, and comprehensively describe the image from multiple dimensions. At the same time, it adopts an adaptive weight adjustment mechanism to dynamically adjust the weights of visual and electrical signal features according to the complexity of the image content and the stability of the electrical signal, making the similarity determination more consistent with the actual situation of the image and effectively improving the accuracy and reliability of appearance and trademark image similarity determination.
[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: a method for determining the similarity of appearance and trademark images based on electrical signal analysis, the method comprising the following specific steps:
[0006] Image acquisition and preprocessing: High-resolution equipment is used to acquire images of the appearance and trademark, which are then converted into digital format and processed by grayscale conversion, noise reduction, and normalization to obtain standardized image data;
[0007] Multi-dimensional feature extraction: Visual features of images are extracted using deep learning models to form visual feature vectors, and electrical signal features are extracted by constructing an electrical signal model and using signal processing methods to form electrical signal feature vectors.
[0008] Adaptive weight adjustment: Calculate the image content complexity and electrical signal stability, and convert their quantized values into weight values of visual features and electrical signal features based on a preset functional relationship to achieve dynamic adjustment;
[0009] Feature fusion and composite vector construction: Based on adaptive weights, the visual feature vector and the electrical signal feature vector are fused into a composite feature vector by weighted summation or weighted averaging.
[0010] Similarity measurement and retrieval result output: The comprehensive feature vector similarity parameters between the image to be retrieved and the images in the database are calculated using vector space similarity or distance measurement methods, and a list of similar images is output in order of parameters.
[0011] Furthermore, in the multi-dimensional feature extraction step, for visual feature extraction, a deep learning model containing multiple feature extraction layers and feature compression layers is adopted. Through feature transformation and compression operations between each layer, color distribution, edge contour and texture detail feature information are extracted from the preprocessed image, and this information is integrated into a visual feature vector.
[0012] Furthermore, in the multi-dimensional feature extraction step, for electrical signal feature extraction, an electrical signal model is constructed based on the electrical characteristics of image pixels, treating each pixel as an independent electrical node, and determining the electrical signal transmission relationship between nodes based on the brightness difference and spatial position relationship between pixels; the electrical signal is converted from the time domain to the frequency domain using a signal domain conversion method to obtain the amplitude and phase information of different frequency components, and a multi-scale decomposition method is used to decompose the electrical signal to extract detailed features at different scales, and the above-mentioned electrical signal related information is integrated into an electrical signal feature vector.
[0013] Furthermore, in the adaptive weight adjustment step, the entropy value and gradient change rate of the image are calculated, where the entropy value is used to characterize the richness of image information and the gradient change rate is used to reflect the drastic change of image pixel values. The two are combined to quantify the complexity of image content. The variance and standard deviation of electrical signal features at different time periods or sampling points are calculated. The variance and standard deviation are used to quantify the fluctuation of electrical signals to evaluate the stability of electrical signals. Based on a preset functional relationship, the quantized values of image content complexity and electrical signal stability are converted into weight values of visual features and electrical signal features, realizing the dynamic adjustment of the weights of the two features in similarity determination.
[0014] Furthermore, in the adaptive weight adjustment step, the complexity of the image content is quantified by combining the image's entropy value and gradient change rate, and its calculation formula is as follows: ,in, It is a quantification value of the complexity of image content. It is the entropy value of an image, used to characterize the richness of information in an image. It is the gradient rate of change of the image, reflecting the degree of drastic change in the pixel values of the image. It is the weighting coefficient of entropy in complex quantification, dynamically adjusted according to the type of appearance and trademark image; the fluctuation of the electrical signal is quantified by variance and standard deviation to evaluate the stability of the electrical signal, and its stability formula is: ,in, It is a quantized value of the stability of the electrical signal. It is the standard deviation of the electrical signal characteristics. It is the variance of the electrical signal characteristics. It is the weighting coefficient of standard deviation in stability quantification.
[0015] Furthermore, in the adaptive weight adjustment step, based on a preset functional relationship, the quantized values of image content complexity and electrical signal stability are converted into weight values of visual features and electrical signal features. The weight calculation formula is as follows: ,in, It is the weight of visual features. It is the weight of the electrical signal characteristics. It is an adjustment coefficient used to control the sensitivity of the weights to changes in the complexity of the image content. It is a baseline value for the complexity of image content. It is a quantified value of the complexity of image content.
[0016] Furthermore, in the feature fusion and comprehensive vector construction step, based on the visual feature weights and electrical signal feature weights obtained through adaptive weight adjustment, a weighted fusion operation is performed on the extracted visual feature vector and electrical signal feature vector. The fusion formula is as follows: ,in, It is a constructed comprehensive feature vector. The weights are for visual features and electrical signal features, respectively. It is a visual feature vector. It is an electrical signal feature vector, thus forming a comprehensive feature vector that can comprehensively reflect the visual attributes and electrical signal attributes of an image.
[0017] Furthermore, in the similarity measurement and retrieval result output step, a similarity measurement model is constructed, the similarity parameter between the comprehensive feature vector of the image to be retrieved and the comprehensive feature vectors of each image stored in the image database is calculated, the images in the database are sorted according to the size of the similarity parameter, and a list of images that are similar to the image to be retrieved is output according to the sorting result, thus completing the similarity retrieval and determination process of appearance and trademark images.
[0018] Furthermore, in the similarity measurement and retrieval result output step, a similarity measurement model is constructed, and its formula is as follows: ,in, It is the comprehensive feature vector of the image to be retrieved. Combined with the image feature vectors stored in the image database The similarity parameter between them , Let be the magnitudes of the two eigenvectors, respectively. This is an adjustment coefficient used to balance dot product similarity and distance-based similarity. It is the attenuation coefficient, which controls the degree to which distance affects similarity.
[0019] Compared with existing technologies, this method for determining the similarity of appearance and trademark images based on electrical signal analysis has the following advantages:
[0020] I. This invention uses a combination of visual features and electrical signal features to determine the similarity of appearance and trademark images. Visual features can capture intuitive information such as color, outline, and texture of an image, while electrical signal features reflect the intrinsic attributes of an image from different dimensions based on the electrical properties of image pixels. The two complement each other and can more comprehensively and accurately describe the characteristics of an image, thereby effectively improving the accuracy of retrieval of complex images and making the similarity determination results more reliable.
[0021] Second, this invention employs an adaptive weight adjustment mechanism to dynamically allocate the weights of visual features and electrical signal features based on the complexity of image content and the stability of electrical signals. The complexity of image content is quantified by entropy and gradient change rate, and the stability of electrical signals is evaluated by variance and standard deviation. The weights are dynamically adjusted based on a preset functional relationship, avoiding the one-sidedness caused by fixed or manually set weights in traditional technologies. It can flexibly adjust feature weights according to the actual situation of different images, making similarity determination more in line with image characteristics, enhancing the comprehensiveness and flexibility of determination, and further improving the accuracy of appearance and trademark image similarity determination.
[0022] Other advantages, objectives and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination or study, or may be learned from the practice of the invention. Attached Figure Description
[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.
[0024] Figure 1 This is a flowchart illustrating the similarity determination method for appearance and trademark image retrieval based on electrical signal analysis.
[0025] Figure 2 This is a flowchart illustrating the multi-dimensional feature extraction process for a similarity determination method for appearance and trademark images based on electrical signal analysis.
[0026] Figure 3 This is a flowchart of an adaptive weight adjustment method for appearance and trademark image retrieval similarity determination based on electrical signal analysis. Detailed Implementation
[0027] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.
[0028] Example 1
[0029] For the determination of trademark image similarity, such as Figure 1As shown, an industrial camera with a resolution of 2048×1536 was used to photograph a trademark. During the shooting process, uniform lighting was ensured to avoid affecting subsequent processing due to light and shadow deviation. Finally, a clear RGB format digital image was obtained. The color image was converted into a grayscale image and noise reduction was performed using a 3×3 filter kernel. Then, the processed grayscale image was normalized to the numerical range of [0,1] so that trademark images with different brightness levels could be processed under a uniform scale, ensuring the standardization of image data.
[0030] like Figure 2 As shown, in terms of visual feature extraction, a deep learning model containing 5 feature extraction layers and 3 feature compression layers is adopted. This model is trained on a large number of trademark image samples. The preprocessed trademark image is input into the model. The first feature extraction layer focuses on capturing the overall outline of the trademark. Subsequent feature extraction layers gradually extract detailed information. Through feature transformation and compression operations at each level, visual features such as color distribution, edge contours and texture details of the trademark are accurately extracted and integrated to form a visual feature vector of length 1024. In terms of electrical signal feature extraction, each pixel of the trademark image is regarded as an electrical node. The electrical signal transmission relationship between nodes is determined based on the brightness difference and spatial position relationship of adjacent pixels. An electrical signal model that can reflect the electrical characteristics of the trademark image is constructed. The electrical signal is converted from the time domain to the frequency domain through the signal domain conversion method to obtain the amplitude and phase information of different frequency components. This information can reflect the electrical characteristics of different lines and color blocks in the trademark pattern. At the same time, the electrical signal is decomposed by the multi-scale decomposition method to extract detailed features at different scales. Finally, it is integrated to form an electrical signal feature vector of length 256.
[0031] like Figure 3 As shown, the entropy and gradient rate of change of the trademark image are calculated. A higher entropy indicates that the trademark contains richer elements such as text and graphics, while a larger gradient rate of change indicates that the line transitions and color transitions in the trademark are more obvious. Combining these two indicators, the complexity of the image content is quantified. The variance and standard deviation of the electrical signal features at different sampling points are calculated. The smaller the variance and standard deviation, the more stable the electrical signal is during transmission, and the more reliable the features are. Based on a preset functional relationship, the quantified values of image content complexity and electrical signal stability are converted into visual feature weights and electrical signal feature weights. The weight calculation formula is as follows: ,in, It is the weight of visual features. It is the weight of the electrical signal characteristics. It is an adjustment coefficient used to control the sensitivity of the weights to changes in the complexity of the image content. It is a baseline value for the complexity of image content. It is a quantitative value of the complexity of the image content. Since the visual elements of the trademark pattern are relatively prominent, the final visual feature weight is 0.65 and the electrical signal feature weight is 0.35.
[0032] Based on the obtained visual feature weight of 0.65 and electrical signal feature weight of 0.35, the visual feature vector and electrical signal feature vector are weighted and fused. The fusion formula is as follows: ,in, It is a constructed comprehensive feature vector. The weights are for visual features and electrical signal features, respectively. It is a visual feature vector. It is an electrical signal feature vector, thus forming a comprehensive feature vector that can comprehensively reflect the visual attributes and electrical signal attributes of the image. During the fusion process, it is ensured that the information of the two features complements each other rather than conflicts, forming a comprehensive feature vector with a length of 1280. This vector not only retains the trademark's distinctive visual identity, but also contains its potential electrical signal characteristics, thus integrating the visual and electrical signal features of the trademark image.
[0033] The trademark image database pre-stores the comprehensive feature vectors of each trademark image obtained through the above processing. The database covers trademarks from different industries and styles. The comprehensive feature vector of the trademark image to be retrieved is compared one by one with the comprehensive feature vectors of all trademark images in the database. The similarity parameter is calculated using the vector space similarity method, and the calculation formula is as follows: ,in, It is the comprehensive feature vector of the image to be retrieved. Combined with the image feature vectors stored in the image database The similarity parameter between them , Let be the magnitudes of the two eigenvectors, respectively. This is an adjustment coefficient used to balance dot product similarity and distance-based similarity. The attenuation coefficient controls the degree of influence of distance on similarity. This method can effectively measure the directional consistency of two vectors in space. Based on the magnitude of the similarity parameter, the trademark images in the database are sorted, and the top 10 trademark images with the highest similarity are selected. These images are most similar to the trademark to be searched in terms of overall style and core elements, thus completing the similarity retrieval and determination of the trademark image.
[0034] Example 2
[0035] For product appearance image similarity determination, a high-precision 3D scanner is used to perform a full-range scan of the product's appearance to ensure that the fine texture of the product's surface can be captured. After obtaining the three-dimensional model, it is converted into a two-dimensional RGB image using professional software. The two-dimensional color image is then converted into a grayscale image, and a 5×5 filter window is used for noise reduction to effectively eliminate noise interference in the image. Subsequently, the processed grayscale image is normalized so that the brightness, contrast and other parameters of the product appearance image are at a unified benchmark, thereby achieving standardization of image data.
[0036] For visual feature extraction, a pre-trained deep learning model containing multiple feature extraction and compression layers is used. This model is trained on a large-scale product appearance dataset and has good extraction capabilities for product morphology, structure, and other features. The pre-processed product appearance image is input into the model, which extracts visual features such as color distribution, edge contours, and texture details of the product appearance step by step from the whole to the local through progressive processing. These features are integrated to form a visual feature vector of length 2048. For electrical signal feature extraction, the pixels of the product appearance image are regarded as electrical nodes. An electrical signal model is constructed based on the brightness differences and spatial relationships between pixels. The model can reflect the differences in electrical conduction in different material areas of the product surface. The frequency domain features of the electrical signal are obtained through signal domain transformation. These features are related to the material properties of the product surface. A multi-scale decomposition method is used to extract detailed features of the electrical signal at different scales, which can capture subtle concavity and convexity changes in the product appearance. These features are integrated to form an electrical signal feature vector of length 384.
[0037] The entropy and gradient rate of change of the product's appearance image are calculated. The entropy reflects the complexity of the product's appearance, such as whether multiple materials are spliced together or whether there are complex patterns. The gradient rate of change reflects the undulations of the product's surface. Combining the two, the complexity of the image content is quantified. The variance and standard deviation of the electrical signal features at different time periods are calculated. Since the material of the product's appearance is relatively stable, the electrical signal fluctuation is small, and a high stability is evaluated. According to the preset functional relationship, the quantified values of image content complexity and electrical signal stability are converted into visual feature weights and electrical signal feature weights. Considering that the electrical signal features can well reflect the product's material, the final visual feature weight is 0.55, and the electrical signal feature weight is 0.45.
[0038] The visual feature vector and the electrical signal feature vector are weighted and averaged with a weight of 0.55 for visual features and 0.45 for electrical signal features. The contribution of the two features is balanced during the fusion process, resulting in a comprehensive feature vector of length 2432. This vector contains both the appearance and shape information of the product and the electrical properties of the material, thus comprehensively reflecting the visual and electrical attributes of the product's appearance.
[0039] The product appearance image database stores comprehensive feature vectors of appearance images, covering multiple fields such as home appliances, machinery, and consumer goods. The comprehensive feature vector of the appearance image of the product to be retrieved is compared with the comprehensive feature vectors of all images in the database, and a similarity parameter is calculated. This method can intuitively reflect the degree of difference between the two feature vectors. Based on the similarity parameter, the product appearance images in the database are sorted, and the top 15 images with the smallest distance are output. These images are most similar to the product to be retrieved in terms of product shape, material performance, etc., thus completing the similarity retrieval and determination of the product appearance image.
[0040] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any simple modifications, equivalent changes and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.
Claims
1. A method for determining the similarity of appearance and trademark images based on electrical signal analysis, characterized in that, The method includes: Image acquisition and preprocessing: High-resolution equipment is used to acquire images of the appearance and trademark, which are then converted into digital format and processed by grayscale conversion, noise reduction, and normalization to obtain standardized image data; Multi-dimensional feature extraction: Visual features of images are extracted using deep learning models to form visual feature vectors, and electrical signal features are extracted by constructing an electrical signal model and using signal processing methods to form electrical signal feature vectors. Adaptive weight adjustment: Calculate the image content complexity and electrical signal stability, and convert their quantized values into weight values of visual features and electrical signal features based on a preset functional relationship to achieve dynamic adjustment; Feature fusion and composite vector construction: Based on adaptive weights, the visual feature vector and the electrical signal feature vector are fused into a composite feature vector by weighted summation or weighted averaging. Similarity measurement and retrieval result output: The comprehensive feature vector similarity parameters between the image to be retrieved and the images in the database are calculated using vector space similarity or distance measurement methods, and a list of similar images is output in order of parameters.
2. The method for determining the similarity of appearance and trademark images based on electrical signal analysis according to claim 1, characterized in that, In the multi-dimensional feature extraction step, for visual feature extraction, a deep learning model containing multiple feature extraction layers and feature compression layers is used. Through feature transformation and compression operations between each layer, color distribution, edge contour and texture detail feature information are extracted from the preprocessed image, and this information is integrated into a visual feature vector.
3. The method for determining the similarity of appearance and trademark images based on electrical signal analysis according to claim 1, characterized in that, In the multi-dimensional feature extraction step, for electrical signal feature extraction, an electrical signal model is constructed based on the electrical characteristics of image pixels, treating each pixel as an independent electrical node. The electrical signal transmission relationship between nodes is determined based on the brightness difference and spatial position relationship between pixels. The electrical signal is converted from the time domain to the frequency domain using a signal domain conversion method to obtain the amplitude and phase information of different frequency components. At the same time, a multi-scale decomposition method is used to decompose the electrical signal, extract detailed features at different scales, and integrate the above-mentioned electrical signal related information into an electrical signal feature vector.
4. The method for determining the similarity of appearance and trademark images based on electrical signal analysis according to claim 1, characterized in that, In the adaptive weight adjustment step, the entropy value and gradient change rate of the image are calculated. The entropy value is used to characterize the richness of image information, and the gradient change rate is used to reflect the drastic change of image pixel values. The two are combined to quantify the complexity of image content. The variance and standard deviation of electrical signal features are calculated at different time periods or sampling points. The fluctuation of the electrical signal is quantified by the variance and standard deviation to evaluate the stability of the electrical signal. Based on a preset functional relationship, the quantized values of image content complexity and electrical signal stability are converted into weight values of visual features and electrical signal features, so as to realize the dynamic adjustment of the weight of the two features in similarity determination.
5. The method for determining the similarity of appearance and trademark images based on electrical signal analysis according to claim 4, characterized in that, In the adaptive weight adjustment step, the complexity of the image content is quantified by combining the image's entropy value and gradient change rate. The calculation formula is as follows: ,in, It is a quantification value of the complexity of image content. It is the entropy value of an image, used to characterize the richness of information in an image. It is the gradient rate of change of the image, reflecting the degree of drastic change in the pixel values of the image. It is the weighting coefficient of entropy in complex quantification, dynamically adjusted according to the type of appearance and trademark image; the fluctuation of the electrical signal is quantified by variance and standard deviation to evaluate the stability of the electrical signal, and its stability formula is: ,in, It is a quantized value of the stability of the electrical signal. It is the standard deviation of the electrical signal characteristics. It is the variance of the electrical signal characteristics. It is the weighting coefficient of standard deviation in stability quantification.
6. The method for determining the similarity of appearance and trademark images based on electrical signal analysis according to claim 5, characterized in that, In the adaptive weight adjustment step, based on a preset functional relationship, the quantized values of image content complexity and electrical signal stability are converted into weight values of visual features and electrical signal features. The weight calculation formula is as follows: ,in, It is the weight of visual features. It is the weight of the electrical signal characteristics. It is an adjustment coefficient used to control the sensitivity of the weights to changes in the complexity of the image content. It is a baseline value for the complexity of image content. It is a quantified value of the complexity of image content.
7. The method for determining the similarity of appearance and trademark images based on electrical signal analysis according to claim 6, characterized in that, In the feature fusion and comprehensive vector construction step, based on the visual feature weights and electrical signal feature weights obtained through adaptive weight adjustment, a weighted fusion operation is performed on the extracted visual feature vectors and electrical signal feature vectors. The fusion formula is as follows: ,in, It is a constructed comprehensive feature vector. The weights are for visual features and electrical signal features, respectively. It is a visual feature vector. It is an electrical signal feature vector, thus forming a comprehensive feature vector that can comprehensively reflect the visual attributes and electrical signal attributes of an image.
8. The method for determining the similarity of appearance and trademark images based on electrical signal analysis according to claim 1, characterized in that, In the similarity measurement and retrieval result output steps, a similarity measurement model is constructed, and the similarity parameter between the comprehensive feature vector of the image to be retrieved and the comprehensive feature vectors of each image stored in the image database is calculated. The images in the database are sorted according to the size of the similarity parameter, and a list of images that are similar to the image to be retrieved is output according to the sorting result, thus completing the similarity retrieval and determination process of appearance and trademark images.
9. The method for determining the similarity of appearance and trademark images based on electrical signal analysis according to claim 8, characterized in that, In the steps of similarity measurement and retrieval result output, a similarity measurement model is constructed, and its formula is as follows: ,in, It is the comprehensive feature vector of the image to be retrieved. Combined with the image feature vectors stored in the image database The similarity parameter between them , Let be the magnitudes of the two eigenvectors, respectively. This is an adjustment coefficient used to balance dot product similarity and distance-based similarity. It is the attenuation coefficient, which controls the degree to which distance affects similarity.