Enhancement method and system for irregular tactile graphic data
Through rectangular and cross filling and random migration processing, the problem of missing edge information and easy confusion in irregular haptic graph data is solved, and the data quality and model performance are significantly improved. It is suitable for a variety of physiological signals and achieves more efficient Perception task completion.
Patent Information
- Application Number
- CN202410985887.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2044-07-23
AI Technical Summary
The prior art has problems of confusion caused by missing edge information and low resolution sensing when processing irregular haptic graphic data, and conventional visual data enhancement methods cannot effectively improve the quality of haptic data, especially due to the lack of robust haptic signals. sex.
The tactile graphic data is filled with rectangular fill and cross fill methods, and the information volume and diversity of the data are increased through random migration processing, singular sample data is eliminated, and standardized processing is carried out to improve data quality.
Improve the amount and diversity of tactile graphic data, improve the perceptual performance of the model, reduce costs, and is suitable for not only tactile data, but also for other physiological signals such as EEG signals and sole pressure signals, significantly improving perception. The quality and accuracy of the task.
Smart Images

Figure CN118967538A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to data processing technology, in particular to an enhancement method and system for irregular tactile graphic data. Background Art
[0002] In the current field of tactile perception, it has become a trend to use high-dimensional sensors to capture irregular graphical sensor data and process it with visual methods, especially in the processing of hand and plantar pressure data. However, the results of some studies show that the cross-application of visual models on physiological signals may not necessarily improve performance. This is mainly due to two aspects: one is the gap in resolution and data volume between the two types of data. This gap will make it difficult for the visual model to find the right complexity to adapt to the physiological data, which will ultimately affect the task performance; the other is the limitations of the data acquisition process. For example, most of the existing high-resolution sensors do not have high sensitivity, resulting in a certain loss of edge information, and the readout circuit is also prone to row and column crosstalk at high resolution. These will reduce the amount of effective information in the data and form a vicious circle with the first problem.
[0003] In view of these situations, most scholars currently tend to improve the accuracy by improving the visual model, but this is costly. The present invention turns to data enhancement, increasing the amount of information and randomness of the data and improving the quality of tactile data from the perspective of changing the original distribution of the data.
[0004] However, the current mainstream data enhancement methods are all based on vision. Conventional enhancement operations include but are not limited to: flipping, cropping, adding noise, and masking. And these conventional enhancement methods require that the data itself has a certain robustness. Tactile signals are physiological signals. Due to their limited dimensions, they do not have strong robustness. Therefore, when these enhancement methods are applied to tactile data, a satisfactory augmentation effect is often not obtained, and may even disrupt the normal distribution of the data in the end. However, there is a certain research gap in the enhancement methods for physiological signals, especially tactile signals; for this reason, the present invention specifically proposes an enhancement scheme for tactile data. Summary of the invention
[0005] In view of the defects and shortcomings of the prior art, the present invention proposes an enhancement method and system for irregular tactile graphic data, which solves the problem of missing edge information that is prone to occur during the acquisition process of tactile graphic data and the confusion caused by low-resolution sensing, increases the amount of information and diversity of tactile graphic data, and improves the perceptual performance of the model.
[0006] The technical solution of the enhancement method adopted in the embodiment of the present invention is: an enhancement method for irregular tactile graphic data, comprising the following steps:
[0007] S1, eliminating the singular sample data in the tactile graphic data and performing standardization processing on the tactile graphic data;
[0008] S2, filling the tactile graphic data with rectangle filling or cross filling;
[0009] Rectangular filling determines the receptive field of the rectangle with the pixel to be filled as the center, performs nonlinear projection after weighted summation of all valid pixel values in the receptive field, and uses the projection result as the pixel value of the central pixel. When one filling is completed, the central pixel will be regarded as a valid pixel, and the pixel core will be transferred to the next pixel to continue the next filling.
[0010] The cross filling determines the receptive field with the filled pixel as the center. The receptive field is all pixels in the globally corresponding rows and columns of the entire tactile graphic data. All valid pixel values in the receptive field are weighted summed and then nonlinearly projected, and the projection result is used as the pixel value of the central pixel.
[0011] S3. Perform random migration processing on the filled tactile graphic data to obtain enhanced tactile graphic data.
[0012] Preferably, the pixel value of the central pixel point of the rectangle filling in step S2 is calculated by the formula:
[0013]
[0014] Where k is the preset receptive field size; valid represents the number of valid pixels within the receptive field; (i, j) represents the relative coordinates of a pixel within the receptive field, X(i, j) represents the pixel value of the central pixel obtained during filling, and F(i, j) represents the pixel value corresponding to the pixel (i, j) in the receptive field; M(i, j) represents whether the pixel (i, j) is a valid pixel, if so, the point is "1", otherwise it is "0"; F represents the matrix of pixels within the receptive field; M represents the discriminant matrix of whether each pixel within the receptive field is a valid pixel; the sigmoid function represents projecting the sum value between 0 and 1.
[0015] Preferably, the pixel value of the center pixel of the cross filling in step S2 is calculated by the formula:
[0016]
[0017] Where X(i,j) represents the pixel value of the central pixel obtained during filling; X(i,a) represents the pixel in the i-th row and a-th column within the receptive field, and X(b,j) represents the pixel in the b-th row and j-th column within the receptive field; the sigmoid function represents projecting the summed value between 0 and 1.
[0018] Preferably, the random migration processing in step S3 adopts the idea of convex combination to splice the cut parts of the two tactile graphic data sample inputs, takes the first tactile graphic data sample input as the enhanced sample object, and cuts out a rectangle from the first tactile graphic data sample input, and convexly combines the original data therein and the data part of the same shape randomly selected from the second tactile graphic data sample input after the same type of data filling processing, and cuts out a shape of the same size from the randomly selected second tactile graphic data sample input and integrates it into the first tactile graphic data sample input.
[0019] Furthermore, the process of random migration processing in step S3 is expressed as:
[0020] x′ a =U⊙x a +V⊙x b
[0021]
[0022] where x a and x b is the tactile graphic data sample input after the same type of data filling processing; U and V are defined to represent the corresponding binary masks, that is, to extract x a 、x b The positions to be deleted, retained or weighted fused between the two tactile graphics data sample inputs, U∈{1,p} 32×32 ,V∈{0,1-p} 32×32 ; p as x a 、x b The coefficient of the fusion of the cut parts of the two tactile graphic data samples input; ⊙ represents element multiplication, ||·|| represents the shape of 1 in the selection matrix, Indicates that the shapes are exactly equal.
[0023] The enhancement system technical solution adopted in the embodiment of the present invention is: an enhancement system for irregular tactile graphic data, comprising the following modules:
[0024] A preprocessing module, used to eliminate singular sample data in the tactile graphic data and perform standardization processing on the tactile graphic data;
[0025] A data filling module, used for filling the tactile graphic data with rectangular filling or cross filling;
[0026] Rectangular filling determines the receptive field of the rectangle with the pixel to be filled as the center, performs nonlinear projection after weighted summation of all valid pixel values in the receptive field, and uses the projection result as the pixel value of the central pixel. When one filling is completed, the central pixel will be regarded as a valid pixel, and the pixel core will be transferred to the next pixel to continue the next filling.
[0027] The cross filling determines the receptive field with the filled pixel as the center. The receptive field is all pixels in the globally corresponding rows and columns of the entire tactile graphic data. All valid pixel values in the receptive field are weighted summed and then nonlinearly projected, and the projection result is used as the pixel value of the central pixel.
[0028] The random migration module performs random migration processing on the filled tactile graphic data to obtain enhanced tactile graphic data.
[0029] Compared with the prior art, the technical effects achieved by the present invention include:
[0030] 1. The present invention can be easily embedded offline and integrated into network training; after data enhancement, the amount of information in tactile graphic data will be increased, the deep learning model can learn more useful knowledge from it to help classification, and the confusion between different categories will also be alleviated due to the random migration of data parts.
[0031] 2. The present invention can improve the quality of tactile perception images and help the model to better complete perception tasks; compared with the prior art of improving task performance by improving the model, the present invention has lower costs, is simpler and more intuitive, and has more obvious effects.
[0032] 3. The present invention is not only applicable to irregular tactile graphic data, but can also be easily applied to other physiological signals, such as EEG signals, plantar pressure signals, etc., meeting the versatility of the method in a wide range and having certain application value. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] Figure 1 is a flow chart of a method for enhancing irregular tactile graphic data in an embodiment of the present invention;
[0034] Figure 2 is a schematic diagram of the tactile graphics data format;
[0035] Figure 3 Schematic diagrams of data filling, where (a) is a schematic diagram of rectangular filling, and (b) is a schematic diagram of cross filling;
[0036] Figure 4 is a schematic diagram of random migration processing. DETAILED DESCRIPTION
[0037] The present invention is further described in detail below in conjunction with embodiments and drawings, but the embodiments of the present invention are not limited thereto.
[0038] Example
[0039] This embodiment provides a method for enhancing irregular tactile graphic data, such as Figure 1 As shown in FIG. 1 , the method can normalize the collected tactile graphic data, add more information to the original tactile graphic data while increasing the differences between different categories, solve the limitations of the front-end collection process from the back-end data processing part, and improve the perception performance. Specifically, the whole method can include the following steps:
[0040] S1. Preprocessing the tactile graphic data.
[0041] The tactile graphic data can be collected by a sensing unit (e.g., a sensor). Let the tactile graphic data collected by the sensor be And assume that it is Figure 2 In order to initially improve the quality and accuracy of tactile graphic data, it is necessary to preprocess it first. The preprocessing is mainly divided into two steps:
[0042] S11. Eliminate singular sample data in the tactile graphic data.
[0043] First, the theoretical upper threshold x of tactile graphics data is defined globally. u , lower threshold x l ; Then traverse all tactile graphic data, if there is a perception unit return value exceeding the upper threshold x in any tactile graphic data x u , the image frame containing the tactile graphic data is determined to be an invalid frame. This phenomenon generally corresponds to the perception unit contacting an abnormally sharp object. If it is retained, its excessive noise will affect the normal distribution of the entire data in subsequent processing. At the same time, if all values in any tactile graphic data x are less than the lower limit threshold x l When the image frame containing the tactile graphic data is detected, it will be determined that the image frame is a non-contact frame and does not contain any information, and is thus determined to be an invalid frame.
[0044] S12, standardize the tactile graphic data x to obtain standardized data x norm The formula for standardization is:
[0045]
[0046] In order to maintain the physical meaning of the information in the data, the standardization of the tactile graphic data will be performed within the preset acquisition time, that is, the tactile graphic data of different acquisition objects will be standardized separately.
[0047] This embodiment defines the collection process of different collection objects as different collection rounds. In each collection round, x ave 、x max and x min Respectively represent the average, maximum, and minimum values of the tactile graphics data subset in this acquisition round. The Clip(·) function is used to limit the value in brackets to the range of [0,1]. This is done because the return value of the perception unit generally has a relatively large threshold range. After standardization, different tactile graphics data will have the same dimension and dimensional unit, so that the return values of all perception units are in the same order of magnitude, and at the same time, it can accelerate the convergence speed of model training.
[0048] S2. Filling the tactile graphic data.
[0049] When high-resolution sensor data applied to human tactile perception is processed by convolutional neural networks or other methods in the field of images, its input is often required to be a regular rectangular structure. At this time, in order to ensure the integrity of this rectangular structure, the existing technology will fill "0" elements in the position where there is actually no sensor. For most algorithms, these zero elements used for filling have no effect on the uniformity of all data. If these invalid zero elements can be supplemented with effective information, it is bound to be able to expand the feature dimension of the original input, and help the model better distinguish different categories of data from another perspective. Solutions have been proposed in previous studies for the problem of missing original image areas, specifically filling the missing input area with salient areas of other categories of images according to certain coefficients, and fusing them according to certain coefficient weights. However, the pixel level of the tactile image is too low, and it is impossible to directly transfer the pixels of other images for filling. This embodiment proposes a new data filling strategy.
[0050] According to the different neighborhood ranges of the pixels to be filled, this embodiment proposes two filling methods. The first is rectangular filling, which is mainly based on the effective information around the blank sensing pixel points; the second is cross filling, which is mainly based on the information of the corresponding row and corresponding column of the sensing unit. The two filling methods will be specifically introduced below.
[0051] 1) Rectangular filling
[0052] When filling the pressure values according to the hand area, there will be many blanks in the upper area of the thumb and the gaps between the other four fingers. These areas can first be filled with data using a method similar to the convolution kernel. Figure 3As shown in sub-figure (a), all dotted grid parts represent a filling receptive field. The rectangular filling determines the receptive field of the rectangle with the pixel to be filled as the center. All valid pixel values in the receptive field are weighted summed and then nonlinearly projected, and the projection result is used as the pixel value of the center pixel. When a filling is completed, the center pixel will be regarded as a valid pixel, and the pixel core will be transferred to the next pixel to continue the next filling. This cycle is repeated according to the mode of formula (2) until all blank pixels are filled.
[0053]
[0054] Where k is the preset receptive field size, the default value is 7; valid represents the number of valid pixels within the receptive field; (i, j) represents the relative coordinates of a pixel within the receptive field, X(i, j) represents the pixel value of the central pixel obtained during filling, and F(i, j) represents the pixel value corresponding to the pixel (i, j) in the receptive field; M(i, j) represents whether the pixel (i, j) is a valid pixel, if so, the point is "1", otherwise it is "0"; F represents the matrix of pixels within the receptive field; M represents the discriminant matrix of whether each pixel within the receptive field is a valid pixel, that is, the valid mask.
[0055] Combined with such a filling method, this embodiment can capture the overall information of different areas within the receptive field, thereby improving the input dimension of the network and the robustness of the network.
[0056] 2) Cross filling
[0057] The principle of cross filling is as follows Figure 3 As shown in the sub-figure (b) of , the receptive field is determined with the filled pixel as the center, and all valid pixel values in the receptive field are weighted summed and then nonlinearly projected, and the projection result is used as the pixel value of the central pixel. Compared with rectangular filling, the receptive field of the central pixel of the cross filling is no longer limited to a rectangle, but the receptive field is extended to all pixels in the globally corresponding rows and columns of the entire tactile graphic data, as expressed in formula (3):
[0058]
[0059] Among them, X(i,j) represents the pixel value of the central pixel point obtained during filling. The pixel neighborhood of this filling method is all the pixels in the i-th row and j-th column, so a and b in the formula will traverse from 0 to 32; X(i,a) represents the pixel point in the i-th row and a-th column within the receptive field, and X(b,j) represents the pixel point in the b-th row and j-th column within the receptive field. After traversal, the information of the corresponding column and row of the filled unit will be mapped to the blank pixel.
[0060] However, if only the pixels within the receptive field are summed, the minimum linearly independent group of the original input vector will not be changed; therefore, this embodiment performs nonlinear projection on the summation result after the summation, and projects the summation value between 0 and 1 based on a sigmoid function. By filling data in this way, a blank element in the finger gap area can capture the information of the corresponding areas of all fingers except the thumb, thereby making full use of broader and global data association information and significantly enhancing the effectiveness of filling. Compared with rectangular filling, the cross filling method has a wider receptive field, and the details that can be extracted are more global.
[0061] In this embodiment, the blank elements in the center of the hand (such as the sides of the webbed fingers and the base of the thumb) are filled in the direction of the surrounding blank elements. After a pixel is filled, it will also be used as a valid value to fill the next pixel. After all the tactile graphic data are filled, the input feature space dimension of the network will be improved to a certain extent, thereby achieving the effect of data enhancement by increasing the amount of data information.
[0062] S3. Perform random migration processing on the filled tactile graphic data to obtain enhanced tactile graphic data.
[0063] When limited perception pixels are used to perceive similar objects, due to the similarity of the objects' shapes, weights and other surface physical properties, the sensor images projected from the specific contact posture surface may not differ to a large extent. If the corresponding tactile frames are used directly for classification at this time, it will be difficult for the model to distinguish the correct results, which will manifest as confusion between similar objects caused by physical limitations. However, since most tactile perceptions have an obvious temporal variation, that is, different surface materials (such as roughness and softness) will act on the tactile interaction, resulting in individual differences in the corresponding perception process. If the contextual information in a perception process can be complemented, the differences in tactile perception frames of similar objects will be amplified, the class spacing can be improved, and the model can better distinguish objects. Based on this, this embodiment proposes random migration processing to help tactile graphic data improve differences and randomness; the process of random migration processing is reflected as follows Figure 4 As shown, the tactile graphic data is enhanced according to formulas (4) and (5).
[0064] x′ a =U⊙x a +V⊙x b (4)
[0065]
[0066] Assume x a and x bis the tactile graphic data sample input after being filled with the same type of data; U and V are defined to represent the corresponding binary masks, that is, to extract x a 、x b The positions of the two inputs that are deleted (weights reset to 0), retained (weights reset to 1), or weighted fused (weights reset to p and 1-p), U∈{1,⊙} 32×32 ,V∈{0,1-p} 32×32 ; p as x a 、x b The coefficients for fusing the two input clipping parts are randomly sampled from the beta distribution; ⊙ represents element-wise multiplication, and ||·|| represents selecting the shape of 1 in the matrix. Indicates that the shapes are exactly equal.
[0067] This embodiment adopts the idea of convex combination instead of directly splicing the cut parts of the two tactile graphic data sample inputs in order to avoid the tactile graphic data from rapidly degenerating and deviating from the data manifold. a As an enhanced sample object, and from the input x a Cut a rectangle from the source and combine the original data with the input x from the same type b The randomly selected data parts of the same shape are convexly combined, and the randomly selected similar input x b Cut off the shape of the same size and integrate it into the input x a Go in.
[0068] The randomness in random migration is mainly reflected in two aspects. First, in the original input x a The size of the selected part is random; second, although the cut parts keep the same shape, the absolute position of the selected area between the two inputs is different. This embodiment achieves the offset effect by a small translation of the starting point. When the input data of different parts of the same category are convexly combined, the input x b The time-frequency information obtained from an image of is randomly shifted to the input x a In another image, a frequency shift occurs. This is the characteristic of the MSDA strategy that the same type of shared time-frequency features are contained in the same category. The closer the data of the same category is, the greater the distance between the categories will be. In addition, when the data set is expanded in this way, the diversity of the data will increase to increase the model's prediction variance for a certain abstract transformation, which can also help the model control its own complexity, alleviate the dependence on the training set data, and generalize to a wider range of tactile data.
[0069] This embodiment also provides an enhancement system for irregular tactile graphic data, including the following modules:
[0070] A preprocessing module, used to eliminate singular sample data in the tactile graphic data and perform standardization processing on the tactile graphic data;
[0071] A data filling module, used for filling the tactile graphic data with rectangular filling or cross filling;
[0072] Rectangular filling determines the receptive field of the rectangle with the pixel to be filled as the center, performs nonlinear projection after weighted summation of all valid pixel values in the receptive field, and uses the projection result as the pixel value of the central pixel. When one filling is completed, the central pixel will be regarded as a valid pixel, and the pixel core will be transferred to the next pixel to continue the next filling.
[0073] The cross filling determines the receptive field with the filled pixel as the center. The receptive field is all pixels in the globally corresponding rows and columns of the entire tactile graphic data. All valid pixel values in the receptive field are weighted summed and then nonlinearly projected, and the projection result is used as the pixel value of the central pixel.
[0074] The random migration module performs random migration processing on the filled tactile graphic data to obtain enhanced tactile graphic data.
[0075] Each module in the enhancement system of this embodiment is used to implement steps S1-S3 respectively, and the detailed implementation process thereof refers to the description of the aforementioned enhancement method.
[0076] The above embodiments are preferred implementation modes of the present invention, but the implementation modes of the present invention are not limited to the above embodiments. Any other changes, modifications, substitutions, combinations, and simplifications that do not deviate from the spirit and principles of the present invention should be equivalent replacement methods and are included in the protection scope of the present invention.
Claims
1. A method for enhancing irregular tactile graphic data, characterized in that: The following steps are involved: S1, eliminating the singular sample data in the tactile graphic data and performing standardization processing on the tactile graphic data; S2, filling the tactile graphic data with rectangle filling or cross filling; Rectangular filling determines the receptive field of the rectangle with the pixel to be filled as the center, performs nonlinear projection after weighted summation of all valid pixel values in the receptive field, and uses the projection result as the pixel value of the center pixel; When one filling is completed, the central pixel will be regarded as a valid pixel, and the pixel core will be transferred to the next pixel to continue the next filling; The cross filling determines the receptive field with the filled pixel as the center. The receptive field is all pixels in the globally corresponding rows and columns of the entire tactile graphic data. All valid pixel values in the receptive field are weighted summed and then nonlinearly projected, and the projection result is used as the pixel value of the central pixel. S3. Perform random migration processing on the filled tactile graphic data to obtain enhanced tactile graphic data.
2. The method for enhancing irregular tactile graphic data according to claim 1, characterized in that: The pixel value of the center pixel of the rectangle filling in step S2 is calculated as follows: Where k is the preset receptive field size; valid represents the number of valid pixels within the receptive field; (i,j) represents the relative coordinates of a pixel within the receptive field, X(i,j) represents the pixel value of the central pixel obtained during filling, F(i,j) represents the pixel value corresponding to the pixel (i,j) in the receptive field; M(i,j) represents whether the pixel (i,j) is a valid pixel. If it is valid, the point is "1", otherwise it is "0"; F represents the matrix of pixels within the receptive field; N represents the discriminant matrix of whether each pixel within the receptive field is a valid pixel; the sigmoid function represents projecting the sum value between 0 and 1.
3. The method for enhancing irregular tactile graphic data according to claim 1, characterized in that: The pixel value of the center pixel of the cross filling in step S2 is calculated as follows: Where X(i,j) represents the pixel value of the central pixel obtained during filling; X(i,a) represents the pixel in the i-th row and a-th column within the receptive field, and X(b,j) represents the pixel in the b-th row and j-th column within the receptive field; the sigmoid function represents projecting the summed value between 0 and 1.
4. The method for enhancing irregular tactile graphic data according to claim 1, characterized in that: In step S3, the random migration process adopts the idea of convex combination to splice the cut parts of the two tactile graphics data sample inputs, takes the first tactile graphics data sample input as the enhanced sample object, and cuts out a rectangle from the first tactile graphics data sample input, and convexly combines the original data therein and the data part of the same shape randomly selected from the second tactile graphics data sample input after the same type of data filling process, and cuts out a shape of the same size from the randomly selected second tactile graphics data sample input and integrates it into the first tactile graphics data sample input.
5. The method for enhancing irregular tactile graphic data according to claim 4, characterized in that: The process of random migration processing in step S3 is expressed as: x′ a =U⊙x a +V⊙x b where x a and x b is the tactile graphic data sample input after the same type of data filling processing; U and V are defined to represent the corresponding binary masks, that is, to extract x a 、x b The positions to be deleted, retained or weighted fused between the two tactile graphics data sample inputs, U∈{1,p} 32×32 ,V∈{0,1-p} 32×32 ; p as x a 、x b The coefficient of the fusion of the cut parts of the two tactile graphic data samples input; ⊙ represents element multiplication, ||·|| represents the shape of 1 in the selection matrix, Indicates that the shapes are exactly equal.
6. The method for enhancing irregular tactile graphic data according to claim 1, characterized in that: The normalization process in step S1 is: where x norm is the standardized data, x is the tactile graphic data; in each acquisition round, x ave 、x max and x min Respectively represent the average, maximum, and minimum values of the tactile graphic data subset in this acquisition round; The Clip(·) function is used to limit the values in brackets to the range of [0,1].
7. An enhancement system for irregular tactile graphic data, characterized in that: Includes the following modules: A preprocessing module, used to eliminate singular sample data in the tactile graphic data and perform standardization processing on the tactile graphic data; A data filling module, used for filling the tactile graphic data with rectangular filling or cross filling; Rectangular filling determines the receptive field of the rectangle with the pixel to be filled as the center, performs nonlinear projection after weighted summation of all valid pixel values in the receptive field, and uses the projection result as the pixel value of the center pixel; When one filling is completed, the central pixel will be regarded as a valid pixel, and the pixel core will be transferred to the next pixel to continue the next filling; The cross filling determines the receptive field with the filled pixel as the center. The receptive field is all pixels in the globally corresponding rows and columns of the entire tactile graphic data. All valid pixel values in the receptive field are weighted summed and then nonlinearly projected, and the projection result is used as the pixel value of the central pixel. The random migration module performs random migration processing on the filled tactile graphic data to obtain enhanced tactile graphic data.
8. The enhancement system for irregular tactile graphic data according to claim 7, characterized in that: In the data filling module, the pixel value of the center pixel of the rectangle filling is calculated as follows: Where k is the preset receptive field size; valid represents the number of valid pixels within the receptive field; (i,j) represents the relative coordinates of a pixel within the receptive field, X(i,j) represents the pixel value of the central pixel obtained during filling, F(i,j) represents the pixel value corresponding to the pixel (i,j) in the receptive field; M(i,j) represents whether the pixel (i,j) is a valid pixel. If so, the value is "1", otherwise it is "0"; F represents the matrix of pixels within the receptive field; M represents the discriminant matrix of whether each pixel within the receptive field is a valid pixel; the sigmoid function represents projecting the sum value between 0 and 1.
9. The enhancement system for irregular tactile graphic data according to claim 7, characterized in that: In the data filling module, the cross fills the pixel value of the central pixel, and the calculation formula is: Where X(i,j) represents the pixel value of the central pixel obtained during filling; X(i,a) represents the pixel in the i-th row and a-th column within the receptive field, and X(b,j) represents the pixel in the b-th row and j-th column within the receptive field; the sigmoid function represents projecting the summed value between 0 and 1.
10. The enhancement system for irregular tactile graphic data according to claim 7, characterized in that: In the random migration module, the idea of convex combination is used to splice the cut parts of the two tactile graphics data sample inputs, the first tactile graphics data sample input is used as the enhanced sample object, and a rectangle is cut out from the first tactile graphics data sample input. The original data and the data part of the same shape randomly selected from the second tactile graphics data sample input after the same type of data filling process are convexly combined, and the shape of the same size is cut out from the randomly selected second tactile graphics data sample input and integrated into the first tactile graphics data sample input.
Citation Information
Patent Citations
Optimization for deconvolution
CN114008664A
Method for improving resolution of touch image of imitated tongue sensor
CN115953326A
Cross-user hand tracking and shape recognition user interface
WO2013166513A2
