Method for generating vision dictionary set by combining different clustering algorithms

Abstract

The invention discloses a method for generating vision dictionary set and relates to the technical fields of mode recognition, computer vision and image understanding. In order to achieve the purpose of capturing different data structures of natural objects and detecting clusters in different shapes and sizes, a framework is required for synthesizing the outputs of a plurality of clustering algorithms, and different clustering algorithms are conducted on a local vision describe subset of a same training image set so as to obtain a vision dictionary set. Different quantized vectors of the same image are obtained on the basis of the vision dictionary set. A classifier set is obtained by learning on different express vector sets of the same training image set. The construction of the vision dictionary set is non-supervised type and a member vision dictionary is independently and concurrently constructed by using different clustering algorithms. A test result shows that the method provided by the invention is capable of obviously increasing the performance of a single vision dictionary, has robustness on background noise and is excellent in recognizing effect.

Description

technical field [0001] The invention belongs to the technical fields of pattern recognition, computer vision and image understanding, and in particular relates to an image classification method based on a visual dictionary. Background technique [0002] The current popular method for image classification is the "bag-of-words" model. Although the "bag-of-words" model is not explicitly shape-modelled, the learned model is effective for irregularly shaped objects or highly structured object classes. After detecting independent salient region blocks and calculating descriptors (that is, feature representations) for these independent blocks, a visual dictionary is obtained by clustering the descriptors of a specific training image set, and then the image is quantized based on the visual dictionary and input into the traditional The classifier gets the classification result. The degree of learning supervision of current image classification methods is generally relatively strong...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the problems of too complex models, too strong supervision and poor robustness in traditional object recognition, the present invention provides a method of combining different clustering algorithms to generate a collective visual dictionary, using visual dictionaries to parallelly utilize existing A variety of information to identify objects