A visual abstraction method of scatter plot based on stippling
Through the dot-tick-based method, the problem of node overlap in the scatter plot is solved, efficient calculation and good visual abstraction effect are achieved, suitable for multi-class scatter plots, and real-time interactive visual enhancement is supported, improving data The efficiency of analysis.
Patent Information
- Application Number
- CN202410581399.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-11
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-05-11
AI Technical Summary
As the data scale grows, the nodes in the scatter plot overlap a large number of times, resulting in over-drawing problems, seriously damaging visual expression and making it difficult to accurately perform data analysis.
Using a scatter plot visual abstract method based on point truncation, through density scalar field construction, adaptive blue noise sampling and interactive visual enhancement, the discrete data points in the scatter plot are converted into continuous density scalar fields, and a polygonal net is used. The grid reshapes the grayscale distribution to realize visual representation and density perception of point truncations.
It realizes efficient calculation and good visual abstraction effect, is suitable for multi-class scatter plots, supports real-time interactive visual enhancement, can quickly switch between different abstract scales, maintain data distribution details, and improve data exploration and visualization efficiency.
Smart Images

Figure CN118502616B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of data processing, and in particular relates to a visual abstraction method of a scatter plot based on stippling. Background Art
[0002] Scatter plot is a simple, intuitive and flexible visualization chart, which is widely used to reveal the data patterns hidden in the data, such as clusters, outliers and trends. Specifically, in the invention patent CN201310443520.5 "A visualization method and system for displaying data based on scatter plot", the data is displayed by using scatter plot, which realizes the visualization of data display in the case of huge data volume, and can be used for data display and data analysis, but the following problems are inevitable:
[0003] As the data scale continues to grow, the nodes in the scatter plot overlap a lot, causing the problem of over-drawing, which seriously damages the visual expression of the scatter plot, making it difficult to accurately perform reliable data analysis on the scatter plot.
[0004] In view of the above technical problems, the present invention provides a visual abstraction method of a scatter plot based on stippling. Summary of the invention
[0005] The purpose of the present invention is to provide a visual abstraction method of a scatter plot based on stippling.
[0006] In order to solve the above technical problems, the present invention provides a scatter plot visual abstraction method based on stippling, which is characterized by specifically comprising:
[0007] Density scalar field construction: Use the kernel density estimation method to transform the discrete two-dimensional data points in the scatter plot into a continuous density scalar field as the input for the next step;
[0008] Adaptive blue noise sampling: Taking the density scalar field as input, the adaptive blue noise sampling algorithm is used to perform sampling in the visual space for the purpose of visually representing the scalar field, constructing a set of sampling points with different spatial frequencies, thereby achieving stippling;
[0009] Sampling point coloring: Each sampling point is colored according to the category of the real data points near its location;
[0010] Interactive visual enhancement: determining the adjustment strategy of the grayscale-density curve of different divided areas according to the mapping relationship between the local density and grayscale of the sampling points, and adjusting the density perception of different divided areas based on the adjustment strategy.
[0011] A further technical solution is to transform the discrete two-dimensional data points in the scatter plot into a continuous density scalar field, specifically including:
[0012] Based on a fine grid composed of a preset area, the scatter plot is divided into different grids based on the density of the scatter points;
[0013] Place a kernel function around each data point of the data sample in each square, and superimpose these kernel functions together to form probability density estimates of different squares through kernel density estimation;
[0014] Different squares are regarded as pixels, and the idea of stippling is used to reshape the grayscale distribution of different squares with polygonal grids of different spatial frequencies, thereby transforming them into a continuous density scalar field.
[0015] A further technical solution is that the specific method of the interactive visual enhancement is:
[0016] The local density of each sampling point is determined by the average distance between each sampling point and its nearest k sampling points;
[0017] An adjustment strategy for the grayscale-density curve of different sampling points is performed based on the local density of different sampling points and the mapping relationship between the local density and the ideal polygonal mesh radius, and the adjustment of the density perception of different sampling points is achieved based on the adjustment strategy.
[0018] A further technical solution is that the value range of k is determined according to the number of sampling points of the scatter diagram, wherein the greater the number of sampling points of the scatter diagram, the greater the value of k.
[0019] A further technical solution is that the specific method of the interactive visual enhancement is:
[0020] Determine the local density of each sampling point by the distance between each sampling point and its nearest k sampling points, and determine the sampling point type of each sampling point by the local density;
[0021] An ideal polygon mesh radius of the sampling point is determined based on the sampling point type of the sampling point, and density perception of different sampling points is adjusted based on the ideal polygon mesh radius.
[0022] 1. Efficient calculation. The present invention implements stippling based on the weightedLBG algorithm. This method divides the space at an exponential rate in the early stage of operation, so it can quickly approach the final convergence state. Compared with the complex data sampling algorithm designed to maintain the relative density of the region, the calculation efficiency is much higher.
[0023] 2. Good visual abstraction effect. The good effect is reflected in three aspects: First, the density distribution can be reconstructed at the perceptual level with high fidelity, and users can intuitively feel the density differences in different regions. This is because the basic goal of stippling is to reconstruct the density field of the original scatter plot; second, it is visually concise and beautiful, thanks to the random, uniform, and non-overlapping distribution of polygonal grids, which is similar to the blue noise distribution of human eye rod cells. Third, rich data distribution details can be retained because small and flexible dots are still used as visual elements. In contrast, data sampling methods cannot avoid overlapping data points, and high-density areas will still overlap to cause visual clutter. The density line method and heat map method usually come at the cost of losing a lot of details and are not suitable for multi-category scatter plots.
[0024] 3. Applicable to multi-class scatter plots. The present invention uses the nearest neighbor data points to color polygonal grids, which is simple and effective. Because polygonal grids are randomly and evenly distributed, this method can be regarded as density adaptive sampling in the category space, and the colored polygonal grids can unbiasedly represent the class distribution of the original scatter plot. In contrast, the data sampling method also requires the design of complex algorithms to maintain the relative density of classes for multi-class data, while the density line method and the density map method are not applicable to multi-class scatter plots because the lines of different classes will cross and the color channels will be occupied.
[0025] 4. The abstract scale can be easily adjusted. The weightedLBG algorithm can control the number of polygonal meshes by adjusting the target quality of the cell. Therefore, the present invention can generate visual abstractions of different scales - the fewer the number of polygonal meshes, the higher the level of abstraction, in order to cope with different visual environments. As shown in Figure xx, the results of the abstract scales from high to low for the same data set are given. At any scale of abstraction, the advantages of the method of the present invention can be maintained. More importantly, weightedLBG is a progressive spatial sampling method. In the early stage of sampling, the number of polygonal meshes changes monotonically. Therefore, the method of the present invention can use the previous sampling results as the initialization between abstractions of different scales without having to start calculations from scratch, so faster switching can be achieved, and this advantage exists for bidirectional switching from high to low and from low to high.
[0026] 5. Support real-time interactive visual enhancement. The curve tool proposed in the present invention can achieve density changes of any degree in any area, so interactive visual enhancement can be achieved, such as enhancing the overall density contrast and enhancing the visibility of outliers. It is particularly important that this interaction is post-processed, and there is no need to repeatedly perform spatial sampling with the interaction, so the real-time nature of the interaction can be guaranteed, which is an important feature that none of the existing methods have. This feature greatly improves the efficiency of exploring data and modifying visualization.
[0027] Other features and advantages will be described in the following description, and partly become apparent from the description, or understood by practicing the invention. The purpose and other advantages of the invention are realized and obtained by the structures particularly pointed out in the description and the drawings.
[0028] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0029] The above and other features and advantages of the present invention will become more apparent by describing in detail exemplary embodiments thereof with reference to the attached drawings.
[0030] Figure 1 It is a flowchart of a stippled scatter plot visual abstraction method.
[0031] Figure 2 It is a flowchart of a method for determining abnormal data labels.
[0032] Figure 3 is a flow chart of a method for determining a sampling point type of a sampling point;
[0033] Figure 4 This is a schematic diagram of the Curves adjustment tool;
[0034] Figure 5 is a schematic diagram of modifying the target quality to create visual abstractions of different abstraction scales;
[0035] Figure 6 is a schematic diagram of the visual enhancement effect. DETAILED DESCRIPTION
[0036] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the embodiments set forth herein; rather, these embodiments are provided so that the present invention will be comprehensive and complete and fully convey the concepts of the example embodiments to those skilled in the art. The same reference numerals in the figures represent the same or similar structures, and thus their detailed description will be omitted.
[0037] The terms "a", "an", "the", and "said" are used to indicate the presence of one or more elements / components / etc.; the terms "comprising" and "having" are used to express an open-ended inclusive meaning and mean that additional elements / components / etc. may be present in addition to the listed elements / components / etc.
[0038] Scatter plots are simple, intuitive, and flexible visualization charts that are widely used to reveal data patterns such as clusters, outliers, and trends. However, as the size of data continues to grow, the nodes in scatter plots overlap a lot, causing the problem of over-drawing, which seriously damages the visual expressiveness of scatter plots. Visual abstraction is a common means to alleviate over-drawing. The idea is to reshape visualization using more abstract and fewer visual elements without losing the original information as much as possible, which reflects the combination of visual design and visual algorithms.
[0039] According to the different spatial dimensions in which the abstraction occurs, scatter plot visual abstraction methods can be divided into two categories: methods that abstract in data space and methods that abstract in visual space.
[0040] The method of abstraction in data space reduces overlap by reducing the number of data points that need to be drawn on the canvas. Typical representatives of this type of method are data point sampling and data point aggregation. The former reduces the data points by extracting a part of the samples from the data point set, but this often leads to the loss of outliers. It is difficult to keep outliers while maintaining the relative density of the region. The latter reduces the data points by aggregating some data points into one point, but this often fails to maintain the relative density of the region. It is difficult to determine the data grouping under the requirement of ensuring that the new points after aggregation do not overlap as much as possible.
[0041] The method of abstraction in visual space reduces overlap by replacing data points with visual elements of a higher level of abstraction. Typical representatives of this type of method are the isodensity line method and the heat map method. The former uses nested closed curves to characterize the density distribution of data points. However, this method loses a lot of information, and because the density curves of different categories cannot avoid crossing, it is not suitable for multi-class scatter plots with extensive overlap between classes. The latter uses squares or hexagonal tiles to replace data nodes, and fills them with different shades of colors to represent the density of the covered area. This method is suitable for very large-scale data, but it is not suitable for multi-class scatter plots because the color channel is occupied.
[0042] The present invention proposes a visual abstraction method for scatter plots based on stippling. The method uses weightedLBG technology to create stippling for the two-dimensional density field of the scatter plot, thereby replacing the original large-scale scatter plot with separated, small-scale polygonal grids to achieve visual abstraction of the scatter plot. The present method can be regarded as a fusion of the method of abstraction in data space and the method of abstraction in visual space. Because, from the visual effect, the present invention does represent the information of the original scatter plot with fewer data points, and from the computational perspective, the present invention is essentially an adaptive blue noise sampling for scalar space.
[0043] Example 1
[0044] In order to solve the above technical problems, Figure 1 As shown, the present invention provides a scatter plot visual abstraction method based on stippling, which is characterized by specifically including:
[0045] Density scalar field construction: Use the kernel density estimation method to transform the discrete two-dimensional data points in the scatter plot into a continuous density scalar field as the input for the next step;
[0046] Specifically, the discrete two-dimensional data points in the scatter plot are transformed into a continuous density scalar field, including:
[0047] Based on a fine grid composed of a preset area, the scatter plot is divided into different grids based on the density of the scatter points;
[0048] Place a kernel function around each data point of the data sample in each square, and superimpose these kernel functions together to form probability density estimates of different squares through kernel density estimation;
[0049] Different squares are regarded as pixels, and the idea of stippling is used to reshape the grayscale distribution of different squares with polygonal grids of different spatial frequencies, thereby transforming them into a continuous density scalar field.
[0050] Furthermore, the formula of the kernel density estimation method is:
[0051]
[0052] in represents the probability density estimate at x, n is the number of sample points, x i is the i-th sample, k() is the kernel function, which uses Gaussian kernel function or Epanechnikov kernel function, and h is the bandwidth parameter, which controls the width of the kernel function.
[0053] Since the density value on each square can be obtained, when the square is regarded as a pixel, the density scalar field is actually equivalent to a grayscale image, and the higher the density, the darker the color. Therefore, the idea of stippling can be used to reshape the grayscale distribution with polygonal grids of different spatial frequencies, that is, more polygonal grids are placed in high-density areas to fill more gray. The present invention uses FastKDE as a method to achieve density scalar field reconstruction, because this method uses FFT (Fast Fourier Transform) to achieve extremely fast speed, and can complete the density estimation of millions of data points on a grid with a resolution of 200*200 in milliseconds.
[0054] Adaptive blue noise sampling: Taking the density scalar field as input, the adaptive blue noise sampling algorithm is used to perform sampling in the visual space for the purpose of visually representing the scalar field, constructing a set of sampling points with different spatial frequencies, thereby achieving stippling;
[0055] Adaptive blue noise sampling is a common method for achieving automatic stippling. The points sampled from space are random and uniform, similar to the polygonal grid distribution of hand-drawn stippling. It is worth emphasizing that adaptive blue noise sampling is different from data point sampling. It is sampled from space according to density probability distribution, and the sampling points do not belong to the original data point set. The adaptive blue noise sampling algorithm used in the present invention is weightedLBG, because the algorithm is computationally efficient and can more carefully and accurately depict the density scalar field when the number of polygonal grids is the same as other algorithms.
[0056] The weightedLBG algorithm is used to transform the density scalar field into a stippled effect, where the number of sampling points is 38,000, which is much smaller than the 4.21 million original data.
[0057] The goal of the weightedLBG algorithm is to construct a Voronoi diagram and take the centroid site of each polygon cell as the sampling point. The basic principle of this algorithm to ensure density distribution reconstruction is to ensure the quality of each polygon cell ∫ A ρ(x)dA (density sum) is equal. This guarantee makes more cells in high-density areas and fewer cells in low-density areas. Strictly speaking, the number of sampling points in any area is proportional to the sum of the density of the scalar field in that area. At the same time, Voronoi partitioning makes the sampling points as far away from each other as possible and do not overlap. The target quality of the cell A s is the only parameter of the weightedLBG algorithm, which controls the scale of visual abstraction. Usually, it is not too small to avoid generating too many sampling points and fail to achieve the purpose of visual abstraction, nor is it too large to make the abstraction level too high and cause the loss of distribution details.
[0058] The algorithm starts with a few random points, constructs a Voronoi partition with these points, and then calculates the mass T of each cell. s , and according to T s and target quality A s , respectively, and perform the following operations on the polygon mesh: Figure 2 The following three operations are shown: delete, move, and split. Then, the next iteration is performed with the current polygon mesh until the quality of all cells is T s Both are close to A s Because the number of cells will grow exponentially in the early stage of iteration, and the acceptable error range ε will gradually become looser with each iteration, the algorithm can converge quickly and achieve efficient calculation.
[0059] It should be noted that the adaptive blue noise sampling algorithm adopts the weightedLBG algorithm.
[0060] Furthermore, a set of sampling points with different spatial frequencies is constructed, specifically including:
[0061] Taking a preset number of random nodes as initialization, and constructing Voronoi partitioning based on the random nodes to obtain different polygonal meshes;
[0062] Calculate the quality of different polygonal meshes, and according to the comparison result between the quality of the polygonal meshes and the target quality, perform three operations of deletion, movement and splitting on the different polygonal meshes to obtain the polygonal meshes after the operation;
[0063] The next round of iteration is performed based on the operated polygonal meshes until the deviation between the quality of all polygonal meshes and the target quality is less than a preset deviation, and a group of sampling points with different spatial frequencies are constructed based on the polygonal meshes.
[0064] Specifically, the calculation formula for the quality of the polygonal mesh is:
[0065] ∫ A ρ(x)dA
[0066] Sampling point coloring: Each sampling point is colored according to the category of the real data points near its location;
[0067] So far, blue noise sampling has completed stippling and achieved visual abstraction of the original scatter plot. However, blue noise sampling can only give the location of the polygonal mesh, but fails to assign colors to the polygonal mesh to represent the category. If the color distribution of the polygonal mesh is consistent with the color distribution of the original data points after coloring, then the above-mentioned visual abstraction method based on stippling can also be applied to multi-class scatter plots.
[0068] In order to achieve a consistent distribution, a simple strategy is to dye the polygonal mesh the color of the data point that is closest to it. Given that the polygonal mesh is randomly and uniformly distributed, on a macro scale, such a coloring scheme actually performs a density adaptive sampling on the category space. In order to quickly find the nearest neighbors of the polygonal mesh, the present invention uses a vp-tree data structure, which allows finding the nearest neighbors of any point in space in a given data set within a time complexity of O(logn), where n is the number of data points. Therefore, if the number of sampling points is m, the time complexity of this step is O(m*logn). The time complexity of constructing a vp-tree is O(nlogn), but this is a one-time calculation and does not need to be calculated in real time in subsequent interactions.
[0069] Specifically, the specific method of dyeing the sampling points is:
[0070] The sampling algorithm of the vp-tree data structure is used to determine the nearest neighbor of the sampling point, and the sampling point is dyed based on the color of the nearest neighbor.
[0071] Furthermore, the specific method for constructing the sampling algorithm of the vp-tree data structure is:
[0072] Obtaining the number of sampling points in the scatter plot, and determining the time complexity based on the number of sampling data points;
[0073] An optimization time threshold is determined according to the time complexity, and a nearest neighbor of the sampling point is determined based on the optimization time threshold.
[0074] Interactive visual enhancement: determining the adjustment strategy of the grayscale-density curve of different divided areas according to the mapping relationship between the local density and grayscale of the sampling points, and adjusting the density perception of different divided areas based on the adjustment strategy.
[0075] The data point size is fixed by default, and the perceived density of the local area depends entirely on the distribution density of the polygonal mesh. However, in reality, the radius of the polygonal mesh will also greatly affect the density perception. For example, after increasing the radius of the polygonal mesh in a certain area, the perceived density of the area will obviously also increase. Based on this principle, the present invention realizes interactive visual enhancement by changing the radius of the polygonal mesh in different density areas, which is used for visual enhancement tasks such as highlighting clusters and enhancing the visibility of outliers.
[0076] In order to achieve fast adjustment for different density areas, such as Figure 4As shown, the present invention designs and implements a curve adjustment tool. The x-axis and y-axis of the curve tool represent the local density and the polygon mesh radius respectively, and the frequency histogram in the middle describes the frequency distribution of the polygon mesh at different local densities. The local density of each polygon mesh is determined by the inverse of the average distance between the polygon mesh and its nearest k polygon meshes. Obviously, the closer the distance to the surrounding polygon meshes, the higher the local density. Using the vp-tree obtained in the previous step, the time complexity of finding the k nearest neighbors of a polygon mesh is O(k*logn). The curve in the middle of the canvas depicts the mapping relationship between local density and the ideal polygon mesh radius. Obviously, initially, since the radius of the polygon meshes is the same, the curve is horizontal. At any time, the user can add control points and move control points in a dotting manner to change the shape of the curve. Then, the polygon mesh radius of each local density is remapped from the x value to the y value described by the curve, thereby realizing the modification of the perceived density. Specifically, it is a curve that depresses the low-density area (reduces the radius) and raises the high-density area (increases the radius), thereby achieving the effect of increasing the density contrast of the entire dot-engraving visualization. The curve is very flexible and can be adjusted by the user to achieve any visual enhancement effect.
[0077] Specifically, the specific method of the interactive visual enhancement is:
[0078] The local density of each sampling point is determined by the average distance between each sampling point and its nearest k sampling points;
[0079] An adjustment strategy for the grayscale-density curve of different sampling points is performed based on the local density of different sampling points and the mapping relationship between the local density and the ideal polygonal mesh radius, and the adjustment of the density perception of different sampling points is achieved based on the adjustment strategy.
[0080] Furthermore, the value range of k is determined according to the number of sampling points of the scatter plot, wherein the greater the number of sampling points of the scatter plot, the greater the value of k.
[0081] Specifically, the specific method of the interactive visual enhancement is:
[0082] Determine the local density of each sampling point by the distance between each sampling point and its nearest k sampling points, and determine the sampling point type of each sampling point by the local density;
[0083] An ideal polygon mesh radius of the sampling point is determined based on the sampling point type of the sampling point, and density perception of different sampling points is adjusted based on the ideal polygon mesh radius.
[0084] Specifically, Figure 3As shown, the method for determining the sampling point type of the sampling point is:
[0085] The k sampling points closest to the sampling point are taken as close-range sampling points, and the close-range sampling points are divided into adjacent sampling points and remote sampling points according to the distance from the sampling point;
[0086] Determine the local proximity density of the adjacent sampling points by the number of the adjacent sampling points and the distances between different adjacent sampling points and the sampling point, and determine the local proximity density of the remote sampling points based on the number of the remote sampling points and the distances between different remote sampling points and the sampling point;
[0087] The local density of the sampling point is determined based on the local neighboring density of the neighboring sampling point and the local neighboring density of the remote sampling point, and the sampling point type of the sampling point is determined according to the local density.
[0088] It should be noted that determining the sampling point type of the sampling point according to the local density specifically includes:
[0089] The local density interval in which the sampling point is located is determined according to the local density of the sampling point, and the sampling point type of the sampling point is determined according to the sampling point type corresponding to the local density interval in which the sampling point is located.
[0090] In another embodiment, the method for determining the sampling point type of the sampling point is:
[0091] The k sampling points closest to the sampling point are taken as close-distance sampling points, and it is determined whether the average distance of the close-distance sampling points is less than a preset distance. If so, the sampling point type of the sampling point is determined based on the preset type. If not, proceed to the next step.
[0092] Dividing the close sampling points into adjacent sampling points and remote sampling points according to the distance from the sampling points, determining whether the number of the adjacent sampling points is greater than the preset number of sampling points, and if so, determining the sampling point type of the sampling point based on the preset type, and if not, proceeding to the next step;
[0093] determining the local proximity density of the adjacent sampling points by the number of the adjacent sampling points and the distances between different adjacent sampling points and the sampling point, and judging whether the local proximity density of the adjacent sampling points meets the requirement; if so, determining the sampling point type of the sampling point based on the preset type; if not, proceeding to the next step;
[0094] The local neighboring density of the remote sampling point is determined based on the number of the remote sampling points and the distances between different remote sampling points and the sampling point, the local density of the sampling point is determined based on the local neighboring density of the neighboring sampling points and the local neighboring density of the remote sampling point, and the sampling point type of the sampling point is determined according to the local density.
[0095] In another embodiment, the method for determining the sampling point type of the sampling point is:
[0096] S11: taking k sampling points closest to the sampling point as close-range sampling points, dividing the close-range sampling points into adjacent sampling points and remote sampling points according to the distance from the sampling point, and judging whether the proportion of the number of adjacent sampling points meets the requirement, if so, proceeding to the next step, if not, determining the sampling point type of the sampling point by a second preset type;
[0097] S12 determines whether the average distance of the adjacent sampling points meets the requirement, if so, determines the sampling point type of the sampling point based on the preset type, if not, proceeds to the next step;
[0098] S13: determining the local proximity density of the adjacent sampling points according to the number of the adjacent sampling points and the distances between different adjacent sampling points and the sampling point, and judging whether the local proximity density of the adjacent sampling points meets the requirement; if so, determining the sampling point type of the sampling point based on the preset type; if not, proceeding to the next step;
[0099] S14 determines the local proximity density of the remote sampling point based on the number of the remote sampling points and the distances between different remote sampling points and the sampling point, determines the local density of the sampling point based on the local proximity density of the adjacent sampling points and the local proximity density of the remote sampling point, and determines the sampling point type of the sampling point according to the local density.
[0100] Based on the above embodiments, the following technical effects are expected to be achieved:
[0101] 1. Efficient calculation. The present invention implements stippling based on the weightedLBG algorithm. This method divides the space at an exponential rate in the early stage of operation, so it can quickly approach the final convergence state. Compared with the complex data sampling algorithm designed to maintain the relative density of the region, the calculation efficiency is much higher.
[0102] 2. Good visual abstraction effect. The good effect is reflected in three aspects: First, the density distribution can be reconstructed at the perceptual level with high fidelity, and users can intuitively feel the density differences in different regions. This is because the basic goal of stippling is to reconstruct the density field of the original scatter plot; second, it is visually concise and beautiful, thanks to the random, uniform, and non-overlapping distribution of polygonal grids, which is similar to the blue noise distribution of human eye rod cells. Third, rich data distribution details can be retained because small and flexible dots are still used as visual elements. In contrast, data sampling methods cannot avoid overlapping data points, and high-density areas will still overlap to cause visual clutter. The density line method and heat map method usually come at the cost of losing a lot of details and are not suitable for multi-category scatter plots.
[0103] 3. Applicable to multi-class scatter plots. The present invention uses the nearest neighbor data points to color polygonal grids, which is simple and effective. Because polygonal grids are randomly and evenly distributed, this method can be regarded as density adaptive sampling in the category space, and the colored polygonal grids can unbiasedly represent the class distribution of the original scatter plot. In contrast, the data sampling method also requires the design of complex algorithms to maintain the relative density of classes for multi-class data, while the density line method and the density map method are not applicable to multi-class scatter plots because the lines of different classes will cross and the color channels will be occupied.
[0104] 4. The abstract scale can be easily adjusted. The weightedLBG algorithm can control the number of polygonal meshes by adjusting the target quality of the cell. Therefore, the present invention can generate visual abstractions of different scales - the fewer the number of polygonal meshes, the higher the level of abstraction, in order to cope with different visual environments. As shown in Figure xx, the results of the abstract scales from high to low for the same data set are shown. At any abstract scale, the advantages of the method of the present invention can be maintained. More importantly, if Figure 5 As shown, a schematic diagram of creating visual abstractions of different abstract scales by modifying the target quality, weightedLBG is a progressive spatial sampling method, and the number of polygonal meshes changes monotonically in the early stage of sampling. Therefore, the method of the present invention can use the previous sampling results as the initialization between abstractions of different scales without having to calculate from scratch, so faster switching can be achieved, and this advantage exists for bidirectional switching from high to low and from low to high.
[0105] 5. Support real-time interactive visual enhancement. The curve tool proposed in the present invention can achieve any degree of density change in any area, so interactive visual enhancement can be achieved, such as enhancing the overall density contrast and enhancing the visibility of outliers. Figure 6As shown in the figure, it is a schematic diagram of the visual enhancement effect. It is particularly important that this interaction is post-processed, and there is no need to repeatedly perform spatial sampling with the interaction, so the real-time nature of the interaction can be guaranteed, which is an important feature that none of the existing methods have. This feature greatly improves the efficiency of exploring data and modifying visualization.
[0106] In the description of this specification, the description of the terms "one embodiment", "a preferred embodiment", etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the embodiment of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.
[0107] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims
1. A visual abstraction method of scatter plot based on stippling, characterized by: Specifically include: Density scalar field construction: Use the kernel density estimation method to transform the discrete two-dimensional data points in the scatter plot into a continuous density scalar field as input for the next step; Adaptive blue noise sampling: Using the density scalar field as input, the adaptive blue noise sampling algorithm is used to visually represent the scalar field. Sampling is performed in the visual space to construct a set of sampling points with different spatial frequencies, thereby achieving stippling. Sampling point coloring: Each sampling point is colored according to the category of the real data points near its location; Interactive visual enhancement: Determine the grayscale-density curve adjustment strategy for different divided areas based on the mapping relationship between the local density and grayscale of the sampling points, and adjust the density perception of different divided areas based on the adjustment strategy; The specific method of the interactive visual enhancement is: The local density of each sampling point is determined by the average distance between each sampling point and its nearest k sampling points; An adjustment strategy for the grayscale-density curve of different sampling points is performed based on the local density of different sampling points and the mapping relationship between the local density and the ideal polygon mesh radius, and the adjustment of the density perception of different sampling points is achieved based on the adjustment strategy.
2. The stippling-based scatter plot visual abstraction method according to claim 1, wherein: Convert the discrete two-dimensional data points in the scatter plot into a continuous density scalar field, including: Based on a fine grid composed of preset areas, the scatter plot is divided into different grids based on the density of the scatter points; Place a kernel function around each data point of the data sample in each square, and superimpose these kernel functions together to form probability density estimates of different squares through kernel density estimation; Different squares are regarded as pixels, and the idea of stippling is used to reshape the grayscale distribution of different squares with polygonal grids of different spatial frequencies, thereby transforming them into a continuous density scalar field.
3. The stippling-based scatter plot visual abstraction method according to claim 2, wherein: The formula of the kernel density estimation method is: in Represents the probability density estimate at x, n is the number of sample points, x i is the i-th sample, K() is the kernel function, which uses Gaussian kernel function or Epanechnikov kernel function, and h is the bandwidth parameter, which controls the width of the kernel function.
4. The stippling-based scatter plot visual abstraction method according to claim 1, wherein: The adaptive blue noise sampling algorithm adopts the weightedLBG algorithm.
5. The stippling-based scatter plot visual abstraction method according to claim 1, wherein: Construct a set of sampling points with different spatial frequencies, including: Taking a preset number of random nodes as initialization, and constructing Voronoi partitioning based on the random nodes to obtain different polygonal meshes; Calculate the quality of different polygon meshes, and perform deletion, movement, and splitting operations on the different polygon meshes based on the comparison results between the quality of the polygon meshes and the target quality to obtain the polygon meshes after the operation; The next iteration is performed based on the operated polygonal mesh until the deviation between the quality of all polygonal meshes and the target quality is less than a preset deviation, and a set of sampling points with different spatial frequencies is constructed based on the polygonal mesh.
6. The stippling-based scatter plot visual abstraction method according to claim 1, wherein: The specific method of dyeing the sampling points is as follows: The sampling algorithm of the vp-tree data structure is used to determine the nearest neighbor of the sampling point, and the sampling point is colored based on the color of the nearest neighbor.
7. The stippling-based scatter plot visual abstraction method according to claim 6, wherein: The specific method for constructing the sampling algorithm of the vp-tree data structure is: Obtaining the number of sampling points in the scatter plot, and determining the time complexity based on the number of sampling data points; An optimization time threshold is determined according to the time complexity, and a nearest neighbor of the sampling point is determined based on the optimization time threshold.
