Classify an image with a multilayer perceptron.

classify_image_class_mlp performs a pixel classification with the multilayer perceptron (MLP) MLPHandle on the multichannel image Image. Before calling classify_image_class_mlp the MLP must be trained with train_class_mlp. Image must have NumInput channels, as specified with create_class_mlp. On output, ClassRegions contains NumOutput regions as the result of the classification. The parameter RejectionThreshold can be used to reject pixels that have an uncertain classification. RejectionThreshold represents a threshold on the probability measure returned by the classification (see classify_class_mlp and evaluate_class_mlp). All pixels having a probability below RejectionThreshold are not assigned to any class. Because an MLP typically assigns pixels that lie outside the convex hull of the training data in the feature space to one of the classes with high probability (confidence), it is useful in many cases to explicitly train a rejection class, even if RejectionThreshold is used, by adding samples for the rejection class with add_samples_image_class_mlp and by re-training the MLP with train_class_mlp.

read_image (Image, 'ic')
gen_rectangle1 (Board, 80, 320, 110, 350)
gen_rectangle1 (Capacitor, 359, 263, 371, 302)
gen_rectangle1 (Resistor, 200, 252, 290, 256)
gen_rectangle1 (IC, 180, 135, 216, 165)
Classes := [Board,Capacitor]
Classes := [Classes,Resistor]
Classes := [Classes,IC]
create_class_mlp (3, 3, 4, 'softmax', 'principal_components', 3, 42,
                  MLPHandle)
add_samples_image_class_mlp (Image, Classes, MLPHandle)
get_sample_num_class_mlp (MLPHandle, NumSamples)
train_class_mlp (MLPHandle, 200, 1, 0.01, Error, ErrorLog)
classify_image_class_mlp (Image, ClassRegions, MLPHandle, 0.5)
dev_display (ClassRegions)
clear_class_mlp (MLPHandle)

If the parameters are valid, the operator classify_image_class_mlp returns the value 2 (H_MSG_TRUE). If necessary an exception handling is raised.

classify_image_class_mlp is reentrant and processed without parallelization.

train_class_mlp, read_class_mlp

class_ndim_box, class_ndim_norm, class_2dim_sup

add_samples_image_class_mlp, create_class_mlp

Foundation