An integrated fast spectrum cluster method

Abstract

The invention discloses a fast spectral clustering method based on integration, input pictures, then define a moving window, make the moving window translate in the horizontal or vertical direction, and divide each input picture into several blocks; Perform color histogram statistics in the HSV color space of the block to extract color feature vectors; use the obtained color feature vectors of each image as the input of spectral clustering to obtain the spectral clustering results of each image, and obtain the label of the corresponding color vector value; use the BIRCH classification tree to classify the color feature vector marked by the label value in step 3; use the BIRCH classification tree result to integrate the results of spectral clustering; mark the integrated label value with different colors to obtain image segmentation the result of. This method corrects the result of spectral clustering through BIRCH classification tree, and the obtained clustering effect is better.

Description

technical field [0001] The invention belongs to the technical field of object recognition, and in particular relates to an integration-based fast spectrum clustering method. Background technique [0002] The problem that is still being solved in the autonomous robot system is the object recognition problem, that is, how to enable the robot perception system to recognize the objects in it, so as to realize the corresponding environment modeling function. Traditional spectral clustering has its own problems: 1) The relevant parameters and similarity matrix construction cannot be determined in a unified way; Both space complexity and space complexity cannot be ignored. When the data size increases, the data cannot be loaded into the memory at one time; 3) The computational complexity of constructing a similarity matrix with a data set size of Q and calculating its eigenvalues ​​and eigenvectors is usually is O(n 3 ), in practical applications, when the order of magnitude exce...

AI Technical Summary

Problems solved by technology

Traditional spectral clustering has its own problems: 1) The relevant parameters and similarity matrix construction cannot be determined in a unified way; Both space complexity and space complexity cannot be ignored. When the data size increases, the data cannot be loaded into the memory at one time; 3) The computational complexity of constructing a similarity matrix with a data set size of Q and calculating its eigenvalues ​​and eigenvectors is usually is O(n 3 ), in practical applications, when the order of magnitude exceeds one thousand, spectral clustering becomes gradually unusable

First, cluster the input data using the K-means algorithm to obtain k clusters, and calculate the cluster centers of all clusters; then establish a correspondence table between the data and the cluster centers; finally use the spectral clustering algorithm to obtain the cluster centers Clustering, get the corresponding label value, and mark the label value to each piece of data through the corresponding table. Although this method speeds up the clustering speed of spectral clustering, it leads to a decrease in clustering accuracy

Images