Indoor scene point cloud semantic segmentation method and system for wearable device

Abstract

The invention relates to the technical field of point cloud data processing, in particular to an indoor scene point cloud semantic segmentation method and system for a wearable device, which can better capture tiny details in a scene through curvature-based feature point selection and local geometric feature extraction, improve the capture capability of fine object edges, and improve the accuracy of point cloud data processing. Therefore, the segmentation precision is obviously improved; meanwhile, by introducing multi-scale local geometric features and self-adaptive neighborhood definition, the neighborhood radius is adjusted in a self-adaptive mode through curvature change, and therefore the method is excellent in performance in a non-uniform point cloud scene; according to the method, the feature extraction process is simplified, and the lightweight multilayer perceptron is used as a classifier, so that the calculation complexity is remarkably reduced. In combination with optimization strategies such as farthest point sampling (FPS) for reducing point cloud processing points, KD tree for accelerating neighborhood search, parallel processing feature extraction and classification and the like, the method can keep real-time performance under the condition of limited computing resources, and the requirements of low-power-consumption equipment are met.

Description

Technical Field

[0001] The present invention relates to the technical field of point cloud data processing, and in particular to a method and system for semantic segmentation of indoor scene point clouds for wearable devices. Background Art

[0002] Indoor scene point cloud semantic segmentation technology in wearable devices is an important direction of 3D data processing, and its goal is to extract meaningful semantic information from sparse point cloud data. The core of current technology is how to effectively process and optimize these data to adapt to hardware limitations and real-time requirements. In recent years, methods that directly process point cloud data have been widely used. PointNet is the first neural network model that directly processes unordered point cloud data. It can extract features from each point and aggregate global features through symmetry functions (such as max-pooling). This method avoids converting point cloud data into voxels or grids, thereby reducing information loss and computational overhead. On this basis, PointNet++ improves on this through a hierarchical sampling mechanism, enhancing the ability to capture local geometric features, making it more suitable for processing complex indoor environments. In addition, a variety of fusion and optimization strategies have emerged in recent years. Models such as JSMNet further improve the accuracy of semantic segmentation by introducing self-attention mechanisms and multi-scale feature fusion, which is particularly suitable for dense indoor point cloud environments. At the same time, some methods combine geometric features with neural networks for joint optimization, effectively improving the segmentation accuracy in complex scenes.

[0003] Although the existing indoor scene point cloud semantic segmentation technology has made significant progress in accuracy and real-time performance, it still has some objective shortcomings: 1. Information loss and lack of details: Although models such as PointNet and PointNet++ are able to process point clouds directly, they still have shortcomings in capturing local details and complex geometric structures, especially when dealing with dense and detailed indoor scenes, which may lead to low segmentation accuracy of fine objects.

[0004] 2. Poor adaptability to uneven point cloud data: Many existing methods perform poorly when faced with uneven point cloud density. For example, in areas of the scene that are far away from the sensor, the point cloud data becomes sparse, resulting in the model being unable to effectively capture the features of these areas.

[0005] 3. High computational overhead and resource consumption: Existing deep learning models often require a lot of computing resources and memory for high-precision segmentation, which is difficult to meet the real-time requirements for resource-limited application scenarios such as wearable devices. Our goal is to reduce the computational complexity of the model by designing a more lightweight network structure while maintaining high segmentation accuracy to meet the needs of low-power devices. Summary of the invention

[0006] In view of this, the present invention proposes a method and system for semantic segmentation of indoor scene point clouds for wearable devices, which are used to reduce information loss, improve adaptability to processing uneven data, and significantly reduce computational overhead while maintaining accuracy. The method is particularly suitable for real-time indoor scene point cloud segmentation tasks in wearable devices.

[0007] In order to achieve the above object, a method for semantic segmentation of indoor scene point cloud for wearable devices is provided, comprising the following steps: S1: Perform downsampling preprocessing on the point cloud data to obtain a downsampled point cloud point set; S2: Selecting feature points from the downsampled point cloud point set to obtain a key point set; S3: extracting features from the key point set; S4: performing feature fusion and classification on the features extracted in S3; S5: Optimize the real-time processing of S1-S4.

[0008] Preferably, in S1, the input point cloud data is pre-processed by downsampling using the farthest point sampling method. Preferably, the farthest point sampling method (down-sampling preprocessing of the input point cloud data) is specifically: The original point cloud data is ,in is the three-dimensional coordinate of the i-th point, set a random number, and start from the random number, select the point farthest from the existing sampling point set in turn until the number of sampling points reaches the set threshold m, so as to obtain the downsampled point cloud point set , where m < n.

[0009] Preferably, in S2, feature point selection based on curvature is used to select feature points of the downsampled point cloud point set, so as to obtain key points, and the key points are screened by local geometric information of curvature.

[0010] Preferably, the calculation formula of the curvature of the point cloud point set is: In the formula, Indicate point The local geometric change rate of each point in the neighborhood and The distance between them represents the geometric difference between the point and the surrounding points; If the curvature of a point exceeds a preset threshold , then the point is selected as a key point to form a key point set ; Among them, the key point set The expression is: .

[0011] Preferably, in S3, feature extraction is performed on the key point set to extract local geometric features and global features of the key point set; the global features include geometric center, color information and global feature formula.

[0012] Preferably, the process of extracting the local geometric features of the key point set is: For each key point , based on its neighborhood Extract local geometric features and use the points in the neighborhood and the center point Geometric relationship capture key points The local features of the key points The local geometric features of are calculated as follows: In the formula, It is the key point and key points The Euclidean distance represents the straight-line distance between two points; Among them, in the key point set In each key point Neighborhood Defined by the radius r, i.e. The point set within the range of radius r is centered: .

[0013] Preferably, the process of extracting the global features of the key point set is: a. Obtaining the geometric center of the key point set; The geometric center of the key point set is the average position of all points in the neighborhood, which represents the center position of the entire neighborhood. The formula is as follows: In the formula, is the geometric center of the key point set, is the number of points in the neighborhood; b. Obtain the color information of the key point set: For each point , extract its color information ; c. Obtain the global feature formula of the key point set: Among them, the global feature of the key point set is a combination of geometric center and color information: The global features of the keypoint set The formula is: .

[0014] Preferably, the feature fusion of the features extracted in S3 is specifically as follows: The local geometric features obtained in S3 are combined with the global features to form the final fusion features. ; The formula is: ; The feature classification of the features extracted by S3 is specifically as follows: A lightweight multi-layer perceptron (MLP) is used as a classifier to perform feature classification on the features extracted by S3.

[0015] According to another aspect of the present invention, a system for semantic segmentation of indoor scene point clouds for wearable devices is provided, wherein the system adopts the above-mentioned method for semantic segmentation of indoor scene point clouds for wearable devices, and the system comprises: A preprocessing module is used to perform downsampling preprocessing operations on the point cloud data to obtain a downsampled point cloud point set; A key point set acquisition module is used to select feature points from the downsampled point cloud point set to obtain a key point set; A feature extraction module, used for extracting features from the key point set; A feature fusion and classification module, used for fusing and classifying the features extracted by S3; The real-time processing optimization module is used to perform real-time processing optimization on the point cloud semantic segmentation process.

[0016] The advantages and beneficial effects of the present invention are: The present invention selects key points with significant geometric features through curvature calculation, and dynamically adjusts the neighborhood size based on the curvature change of each point, so that the model can effectively extract features in dense and sparse areas. This adaptive neighborhood division method not only improves the model's sensitivity to detail-rich areas, but also maintains the integrity of features when processing sparse point clouds; through curvature-based feature point selection and local geometric feature extraction, it can better capture the tiny details in the scene. The extraction process of local geometric features no longer relies on simple global operations, but improves the ability to capture the edges of fine objects by calculating the geometric change rate within the point cloud neighborhood, thereby significantly improving the segmentation accuracy; The present invention innovatively combines local and global information by introducing a lightweight feature extraction strategy that combines local geometric features with global geometric center and color information. This fusion not only maintains high efficiency, but also strikes a balance between global structure and local details, improving the accuracy and robustness of segmentation. The present invention introduces multi-scale local geometric features and adaptive neighborhood definition, and adaptively adjusts the neighborhood radius through curvature changes to ensure that effective features can be captured in both sparse and dense areas, thereby performing well in uneven point cloud scenes. The adaptive mechanism not only improves the model's sensitivity to sparse areas of point cloud data, but also effectively enhances the overall segmentation performance; The present invention significantly reduces the computational complexity by simplifying the feature extraction process and using a lightweight multi-layer perceptron (MLP) as a classifier. Combining the optimization strategies such as farthest point sampling (FPS) to reduce the number of point cloud processing points, KD tree to accelerate neighborhood search, and parallel processing of feature extraction and classification, the method can maintain real-time performance under limited computing resources and meet the needs of low-power devices; Aiming at the real-time processing requirements of wearable devices, this paper specially designs efficient parallel processing and KD tree to accelerate neighborhood search. Compared with traditional linear neighborhood search, the introduction of KD tree reduces the search time complexity to , combined with parallel processing, the overall processing efficiency is significantly improved, and it can perform well in real-time segmentation tasks. This efficient optimization makes the method particularly suitable for wearable device application scenarios that require fast response. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the drawings required for use in the description of the present invention or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 A flowchart of a method for semantic segmentation of indoor scene point cloud for wearable devices provided in an embodiment of the present invention; Figure 2 A schematic diagram of a process from data processing to outputting semantic segmentation results for semantic segmentation of indoor scene point cloud for wearable devices provided by an embodiment of the present invention; Figure 3 A schematic diagram of an indoor scene point cloud semantic segmentation system for wearable devices provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] Attached Figure 1 A flowchart of a method for semantic segmentation of indoor scene point clouds for wearable devices is shown. Figure 2 A schematic diagram of the process from data processing to outputting semantic segmentation results for indoor scene point cloud semantic segmentation of wearable devices is shown in the accompanying figure Figure 1 , Figure 2 As shown, a method for semantic segmentation of indoor scene point cloud for wearable devices comprises the following steps: S1: Perform downsampling preprocessing on the point cloud data to obtain a downsampled point cloud point set; In this step, the farthest point sampling (FPS) method is used to perform downsampling preprocessing on the input point cloud data. The farthest point sampling (FPS) method is a commonly used sampling algorithm, which is mainly used to process point cloud data. Its core idea is to make the distance between all sampling points as far as possible, so as to achieve uniform discrete sampling of data. The principle of the algorithm is as follows: Initial point selection: First, select a starting point from the point cloud, usually the point farthest from the center of gravity of the point cloud is selected as the first sampling point. Iterative sampling: For the remaining points, calculate the distance from each point to the current sampling point, select the point with the farthest distance as the next sampling point, and update the sampling point set. Distance measurement: In point cloud data, Euclidean distance is generally used to measure the distance between points. This algorithm reduces the number of points calculated while maintaining the geometric distribution characteristics of the point cloud.

[0021] Specifically, the farthest point sampling (FPS) method is used to perform downsampling preprocessing on the input point cloud data as follows: The original point cloud data is ,in is the three-dimensional coordinate of the i-th point, set a random number, and start from the random number, select the point farthest from the existing sampling point set in turn until the number of sampling points reaches the set threshold m, so as to obtain the downsampled point cloud point set , where m < n; by adopting the farthest point sampling method, the down-sampled point set is guaranteed to have a uniform distribution and can effectively represent the overall structure of the point cloud.

[0022] S2: Selecting feature points from the downsampled point cloud point set to obtain a key point set; In order to reduce unnecessary calculation amount, this embodiment does not process all point cloud point sets, but selects key points with obvious geometric features for feature extraction. Therefore, in this step, feature point selection based on curvature is used to select feature points of the downsampled point cloud point set to obtain key points, and the key points are screened by local geometric information of curvature; The calculation formula of the curvature of the point cloud point set is: In the formula, Indicate point The local geometric change rate of each point in the neighborhood and The distance between them represents the geometric difference between the point and the surrounding points; If the curvature of a point exceeds a preset threshold , then the point is selected as a key point to form a key point set ; Among them, the key point set The expression is: .

[0023] S3: extracting features from the key point set; In this step, feature extraction is performed on the key point set to extract local geometric features and global features of the key point set; The global features include geometric center, color information and global feature formula; The process of extracting the local geometric features of the key point set is as follows: For each key point , based on its neighborhood Extract local geometric features and use the points in the neighborhood and the center point Geometric relationship capture key points The local features of the key points The local geometric features of are calculated as follows: In the formula, It is the key point and key points The Euclidean distance represents the straight-line distance between two points. By calculating the square of this distance, the influence of farther points can be amplified, making changes in the local geometric structure more obvious; Among them, in the key point set In each key point Neighborhood Defined by the radius r, i.e. The point set within the range of radius r is centered: In this step, local geometric features reflects the distribution of points in the neighborhood. If all neighborhood points are close to , the eigenvalue is small; if the points in the neighborhood are more dispersed, the eigenvalue will increase. This information is very useful for distinguishing different geometric structures and can help the model understand local geometric changes.

[0024] Traditional methods often have difficulty dealing with uneven point cloud data. For example, when the point cloud density is low (such as in areas far from the sensor), it is difficult for the model to effectively segment these areas. This step introduces multi-scale local geometric features and adaptive neighborhood definitions, and adaptively adjusts the neighborhood radius through curvature changes to ensure that effective features can be captured in both sparse and dense areas, thereby performing well in uneven point cloud scenes. The adaptive mechanism not only improves the model's sensitivity to sparse areas of point cloud data, but also effectively enhances the overall segmentation performance.

[0025] The process of extracting the global features of the key point set is as follows: a. Obtaining the geometric center of the key point set; The geometric center of the key point set is the average position of all points in the neighborhood, which represents the center position of the entire neighborhood. The formula is as follows: In the formula, is the geometric center of the key point set, reflecting the spatial position of the entire neighborhood. is the number of points in the neighborhood, ensuring that the geometric center is the weighted average of all neighborhood points; b. Obtain the color information of the key point set: For each point , extract its color information Color can be an important semantic clue in indoor scenes. For example, different furniture or walls can have different colors, which helps to identify different object categories. c. Obtain the global feature formula of the key point set: Among them, the global feature of the key point set is a combination of geometric center and color information: The global features of the keypoint set The formula is: This global feature is simple and efficient, suitable for use in resource-limited environments, while providing additional semantic information through color and geometric position.

[0026] S4: performing feature fusion and classification on the features extracted in S3; The feature fusion of the features extracted by S3 is specifically as follows: The local geometric features obtained in S3 are combined with the global features to form the final fusion features. ; The formula is: ; The feature classification of the features extracted by S3 is specifically as follows: Using a lightweight multi-layer perceptron (MLP) as a classifier to perform feature classification on the features extracted by S3; ‌Multilayer Perceptron (MLP)‌ is a feedforward artificial neural network model. The network layers composed of multiple neurons are connected in a fully connected manner. MLP consists of multiple layers composed of several neurons, including input layer, hidden layer and output layer. The input layer receives input data and passes the data to the hidden layer. The hidden layer converts the input value into output value through the activation function and passes it to the output layer. The output layer gives the final prediction result. MLP optimizes the weights and biases of neurons through training so that the network can learn the mapping relationship between input and output. The back-propagation algorithm is used during training to calculate the error between the predicted value and the actual value, and the error is back-propagated to each neuron in the network to update the weights and biases so that the network can better approximate the target function.‌ In this step, the multilayer perceptron includes two hidden layers, each layer uses the ReLU activation function, and the input of the multilayer perceptron is the fusion feature , the output is a point Semantic category prediction ; The formula is: In this step, the semantic segmentation result of the point cloud is finally output through the lightweight classifier.

[0027] As a preferred embodiment, the method further includes the following steps: S5: Optimize the real-time processing of S1-S4; Wherein, the S5 specifically includes: parallel processing of KD number neighborhood search and neighborhood feature extraction and classification; In order to speed up the neighborhood search process, the KD tree structure is used. The KD tree (K-Dimensional Tree) is a data structure for efficiently processing multidimensional data. It is a special binary tree structure used to organize points in k-dimensional space. The KD tree recursively divides the high-dimensional space into a series of hyper-rectangular regions. Each node represents a hyperplane, which divides the space into two subspaces. This structure can be used in logarithmic time complexity. The nearest neighbor search is completed within the kernel, which can significantly improve the processing efficiency.

[0028] The parallel processing of neighborhood feature extraction and classification is specifically as follows: ,Neighborhood feature extraction and classification can be processed in parallel to further improve the computational efficiency, so that the entire process can run in environments with limited computing resources such as wearable devices.

[0029] Embodiment 2: This embodiment includes an indoor scene point cloud semantic segmentation system for wearable devices, Figure 3 The structure diagram of an indoor scene point cloud semantic segmentation system for wearable devices is shown in the attached figure. Figure 3 As shown, the system adopts an indoor scene point cloud semantic segmentation method for wearable devices in embodiment 1, and the system includes: A preprocessing module is used to perform downsampling preprocessing operations on the point cloud data to obtain a downsampled point cloud point set; A key point set acquisition module is used to select feature points from the downsampled point cloud point set to obtain a key point set; A feature extraction module, used for extracting features from the key point set; A feature fusion and classification module, used for fusing and classifying the features extracted by S3; The real-time processing optimization module is used to perform real-time processing optimization on the point cloud semantic segmentation process.

[0030] Embodiment 3, this embodiment includes a computer-readable storage medium, on which a data processing program is stored, and the data processing program is executed by a processor. Embodiment 1 is a method for semantic segmentation of indoor scene point clouds for wearable devices.

[0031] It should be understood by those skilled in the art that the embodiments of this article can be provided as methods, devices (equipment), or computer program products. Therefore, this article can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Including but not limited to RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassette, tape, disk storage or other magnetic storage device, or any other medium that can be used to store desired information and can be accessed by a computer. In addition, it is well known to those of ordinary skill in the art that communication media generally contain computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0032] This document is described with reference to flowcharts and / or block diagrams of methods, apparatuses (devices), and computer program products according to the embodiments of this document. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0033] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0034] The embodiments and / or implementation methods described above are only used to illustrate the preferred embodiments and / or implementation methods of the technology of the present invention, and are not intended to impose any formal restrictions on the implementation methods of the technology of the present invention. Any person skilled in the art may make some changes or modifications to other equivalent embodiments without departing from the scope of the technical means disclosed in the content of the present invention, but they should still be regarded as the technology or embodiments that are essentially the same as the present invention. Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described here, and that various obvious changes, readjustments and substitutions can be made to those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention is described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may also include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the attached claims.

Claims

1. A method for semantic segmentation of indoor scene point clouds for wearable devices, characterized in that: The method comprises the following steps: S1: Perform downsampling preprocessing on the point cloud data to obtain a downsampled point cloud point set; S2: Selecting feature points from the downsampled point cloud point set to obtain a key point set; S3: extracting features from the key point set; S4: performing feature fusion and classification on the features extracted in S3; S5: Optimize the real-time processing of S1-S4.

2. The method for semantic segmentation of indoor scene point cloud for wearable devices according to claim 1, characterized in that: In S1, the farthest point sampling method is used to perform down-sampling preprocessing on the input point cloud data.

3. The method for semantic segmentation of indoor scene point cloud for wearable devices according to claim 2, characterized in that: The farthest point sampling method (downsampling preprocessing of the input point cloud data) is specifically as follows: The original point cloud data is ,in is the three-dimensional coordinate of the i-th point, set a random number, and start from the random number, select the point farthest from the existing sampling point set in turn until the number of sampling points reaches the set threshold m, so as to obtain the downsampled point cloud point set , where m < n.

4. The method for semantic segmentation of indoor scene point cloud for wearable devices according to claim 1, characterized in that: In S2, feature point selection based on curvature is used to select feature points of the downsampled point cloud point set, so as to obtain key points, and the key points are screened by local geometric information of curvature.

5. The method for semantic segmentation of indoor scene point cloud for wearable devices according to claim 4, characterized in that: The calculation formula of the curvature of the point cloud point set is: In the formula, Indicate point The local geometric change rate of each point in the neighborhood and The distance between them represents the geometric difference between the point and the surrounding points; If the curvature of a point exceeds a preset threshold , then the point is selected as a key point to form a key point set ; Among them, the key point set The expression is: 。 6. The method for semantic segmentation of indoor scene point cloud for wearable devices according to claim 1, characterized in that: In S3, feature extraction is performed on the key point set to extract local geometric features and global features of the key point set; the global features include geometric center, color information and global feature formula.

7. The method for semantic segmentation of indoor scene point cloud for wearable devices according to claim 6, characterized in that: The process of extracting the local geometric features of the key point set is: For each key point , based on its neighborhood Extract local geometric features and use the points in the neighborhood and the center point Geometric relationship capture key points The local features of the key points The local geometric features of are calculated as follows: In the formula, It is the key point and key points The Euclidean distance represents the straight-line distance between two points; Among them, in the key point set In each key point Neighborhood Defined by the radius r, i.e. The point set within the range of radius r is centered: 。 8. The method for semantic segmentation of indoor scene point cloud for wearable devices according to claim 1, characterized in that: The process of extracting the global features of the key point set is: a. Obtaining the geometric center of the key point set; The geometric center of the key point set is the average position of all points in the neighborhood, which represents the center position of the entire neighborhood. The formula is as follows: In the formula, is the geometric center of the key point set, is the number of points in the neighborhood; b. Obtain the color information of the key point set: For each point , extract its color information ; c. Obtain the global feature formula of the key point set: Among them, the global feature of the key point set is a combination of geometric center and color information: The global features of the keypoint set The formula is: 。 9. The method for semantic segmentation of indoor scene point cloud for wearable devices according to claim 8, characterized in that: The feature fusion of the features extracted by S3 is specifically as follows: The local geometric features obtained in S3 are combined with the global features to form the final fusion features. ; The formula is: ； The feature classification of the features extracted by S3 is specifically as follows: A lightweight multi-layer perceptron (MLP) is used as a classifier to perform feature classification on the features extracted by S3.

10. A point cloud semantic segmentation system for indoor scenes for wearable devices, characterized in that: The system adopts the indoor scene point cloud semantic segmentation method for wearable devices according to any one of claims 1 to 9, and the system includes: A preprocessing module is used to perform downsampling preprocessing operations on the point cloud data to obtain a downsampled point cloud point set; A key point set acquisition module is used to select feature points from the downsampled point cloud point set to obtain a key point set; A feature extraction module, used for extracting features from the key point set; A feature fusion and classification module, used for fusing and classifying the features extracted by S3; The real-time processing optimization module is used to perform real-time processing optimization on the point cloud semantic segmentation process.

Images