Spatial nearest neighbor-based image segmentation method with weight constraint

Abstract

The invention relates to the technical field of computer vision, in particular to a spatial nearest neighbor-based image segmentation method with weight constraint, which comprises the following stepsof: defining a color domain, collecting an original image of an open source as a data set, manufacturing a label image, and establishing a deep neural network model integrated with a KD tree; inputting the original image and the label image into an encoder full-connection layer network structure for training, outputting a neuron value, weighting different color components in space according to the neuron value, and realizing nearest neighbor color domain search through a KD tree, thereby realizing image segmentation by utilizing the nearest neighbor color domain; the adopted network model islow in complexity, low in time consumption and easy to train, the image segmentation method is high in universality and low in cost, and a large amount of manpower and time required for selecting andmarking training samples are saved; according to the method, the image segmentation is realized by adopting the spatial nearest neighbor of the color domain, the segmentation is more accurate, the precision is higher, the speed is higher, and the practicability is strong.

Description

technical field [0001] The invention relates to the technical field of computer vision, in particular to a space nearest neighbor-based image segmentation method with weight constraints. Background technique [0002] With the development of machine learning, the accuracy of unsupervised learning methods is getting higher and higher. Compared with supervised learning methods, unsupervised learning does not require labeling and is usually implemented with certain specific rules. As an important tool of image processing, image segmentation is one of the three major tasks of image processing, and it is in a basic but very important position in the complex image processing process. [0003] At present, although the majority of researchers have explored tens of thousands of image segmentation methods, these image segmentation methods lack versatility, and image segmentation methods based on neural networks generally require a large amount of supervised learning for each type of ta...

AI Technical Summary

Problems solved by technology

[0004] The present invention provides a weight-constrained image segmentation method based on spatial nearest neighbors. The technical problem to be solved is that the current image segmentation method lacks versatility, and the supervised learning based on the neural network requires manual labeling, which has high cost and low efficiency.

Images