create_trained_component_model — Prepare a component model for matching based on trained components.
create_trained_component_model( : : ComponentTrainingID, AngleStart, AngleExtent, MinContrastComp, MinScoreComp, NumLevelsComp, AngleStepComp, OptimizationComp, MetricComp, PregenerationComp : ComponentModelID, RootRanking)
create_trained_component_model prepares the training result, which is passed in ComponentTrainingID, as a component model for matching. The output parameter ComponentModelID is a handle for this model, which is used in subsequent calls to find_component_model. In contrast to create_component_model, the model components must have been previously trained using train_model_components before calling create_trained_component_model.
The parameters AngleStart and AngleExtent determine the range of possible rotations of the component model in an image.
Internally, a separate shape model is built for each model component (see create_shape_model). Therefore, the parameters MinContrastComp, MinScoreComp, NumLevelsComp, AngleStepComp, OptimizationComp, MetricComp, and PregenerationComp correspond to the parameters of create_shape_model, with the following differences: First, the parameter Optimization of create_shape_model provides the possibility to reduce the number of model points as well as the possibility to completely pregenerate the shape model. In contrast, the operator create_trained_component_model uses a separate parameter PregenerationComp in order to decide whether the shape models should be completely pregenerated or not. A second difference concerning the parameter MinScoreComp should be noted. When using the shape-based matching, this parameter needs not be passed when preparing the shape model using create_shape_model but only during the search using find_shape_model. In contrast, when preparing the component model it is favorable to analyze rotational symmetries of the model components and similarities between the model components. However, this analysis only leads to meaningful results if the value for MinScoreComp that is used during the search (see find_component_model) is already approximately known. After the search with find_component_model the pose parameters of the components in a search image are returned. Note that the pose parameters refer to the reference points of the components. The reference point of a component is the center of gravity of its associated region that is returned in ModelComponents of train_model_components.
The parameters MinContrastComp, NumLevelsComp, AngleStepComp, and OptimizationComp can be automatically determined by passing 'auto' for the respective parameters.
All component-specific input parameters (parameter names terminate with the suffix Comp) must either contain one element, in which case the parameter is used for all model components, or must contain the same number of elements as the number of model components contained in ComponentTrainingID, in which case each parameter element refers to the corresponding component in ComponentTrainingID.
In addition to the individual shape models, the component model also contains information about the way the single model components must be searched relative to each other using find_component_model in order to minimize the computation time of the search. For this, the components are represented in a tree structure. First, the component that stands at the root of this search tree (root component) is searched. Then, the remaining components are searched relative to the pose of their predecessor in the search tree.
The root component can be passed as an input parameter of find_component_model during the search. To what extent a model component is suited to act as root component depends on several factors. In principle, a model component that can be found in the image with a high probability should be chosen. Therefore, a component that is sometimes occluded to a high degree or that is missing in some cases is not well suited to act as root component. Additionally, the computation time that is associated with the root component during the search can serve as a criterion. A ranking of the model components that is based on the latter criterion is returned in RootRanking. In this parameter the indices of the model components are sorted in descending order according to their associated computation time, i.e., RootRanking contains the index of the model component that, chosen as root component, allows the fastest search. Note that the ranking returned in RootRanking represents only a coarse estimation. Furthermore, the calculation of the root ranking assumes that the image size as well as the value of the system parameter 'border_shape_models' are identical when calling create_trained_component_model and find_component_model.
Handle of the training result.
Smallest rotation of the component model.
Default value: -0.39
Suggested values: -3.14, -1.57, -0.79, -0.39, -0.20, 0.0
Extent of the rotation of the component model.
Default value: 0.79
Suggested values: 6.28, 3.14, 1.57, 0.79, 0.39
Restriction: AngleExtent >= 0
Minimum contrast of the components in the search images.
Default value: 'auto'
Suggested values: 'auto', 10, 20, 20, 40
Restriction: MinContrastComp >= 0
Minimum score of the instances of the components to be found.
Default value: 0.5
Suggested values: 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0
Minimum increment: 0.01
Recommended increment: 0.05
Restriction: (0 <= MinScoreComp) && (MinScoreComp <= 1)
Maximum number of pyramid levels for the components.
Default value: 'auto'
List of values: 'auto', 1, 2, 3, 4, 5, 6, 7, 8, 9, 10
Step length of the angles (resolution) for the components.
Default value: 'auto'
Suggested values: 'auto', 0.0175, 0.0349, 0.0524, 0.0698, 0.0873
Restriction: AngleStepComp >= 0
Kind of optimization for the components.
Default value: 'auto'
List of values: 'auto', 'none', 'point_reduction_low', 'point_reduction_medium', 'point_reduction_high'
Match metric used for the components.
Default value: 'use_polarity'
List of values: 'use_polarity', 'ignore_global_polarity', 'ignore_local_polarity', 'ignore_color_polarity'
Complete pregeneration of the shape models for the components if equal to 'true'.
Default value: 'false'
List of values: 'true', 'false'
Handle of the component model.
Ranking of the model components expressing the suitability to act as the root component.
* Get the model image. read_image (ModelImage, 'model_image.tif') * Define the regions for the initial components. gen_rectangle2 (InitialComponentRegions, 212, 233, 0.62, 167, 29) gen_rectangle2 (Rectangle2, 298, 363, 1.17, 162, 34) gen_rectangle2 (Rectangle3, 63, 444, -0.26, 50, 27) gen_rectangle2 (Rectangle4, 120, 473, 0, 33, 20) concat_obj (InitialComponentRegions, Rectangle2, InitialComponentRegions) concat_obj (InitialComponentRegions, Rectangle3, InitialComponentRegions) concat_obj (InitialComponentRegions, Rectangle4, InitialComponentRegions) * Get the training images. gen_empty_obj (TrainingImages) for i := 1 to 4 by 1 read_image (TrainingImage, 'training_image-'+i+'.tif') concat_obj (TrainingImages, TrainingImage, TrainingImages) endfor * Extract the model components and train the relations. train_model_components (ModelImage, InitialComponentRegions, \ TrainingImages, ModelComponents, 22, 60, 30, 0.65, \ 0, 0, rad(60), 'speed', 'rigidity', 0.2, 0.4, \ ComponentTrainingID) * Create the component model based on the training result. create_trained_component_model (ComponentTrainingID, -rad(30), rad(60), 10, \ 0.5, 'auto', 'auto', 'none', \ 'use_polarity', 'false', ComponentModelID, \ RootRanking) * Find the component model in a run-time image. read_image (SearchImage, 'search_image.tif') find_component_model (SearchImage, ComponentModelID, RootRanking, -rad(30), \ rad(60), 0.5, 0, 0.5, 'stop_search', 'prune_branch', \ 'none', 0.55, 'least_squares', 0, 0.9, ModelStart, \ ModelEnd, Score, RowComp, ColumnComp, AngleComp, \ ScoreComp, ModelComp)
If the parameters are valid, the operator create_trained_component_model returns the value 2 (H_MSG_TRUE). If necessary an exception is raised.