Image classification method based on improved bag-of-visual word model

Abstract

The invention discloses an image classification method based on an improved bag-of-visual word model. The method comprises the following steps: S1, fine-resolution SIFT feature points are extracted; S2, fine-resolution SIFT feature point pairs are extracted; S3, a visual dictionary is built; and S4, image classification is carried out. The method provides a visual phrase building algorithm based on a minimum spanning tree, the minimum spanning tree is built for all fine-resolution SIFT features of each training image, the SIFT feature point pairs are built on the basis, K-means clustering is then carried out, visual phrases containing the space information are thus obtained, the visual words and the visual phrases built by using the above method together build a visual histogram, and image classification is completed through an SVM. The image classification method based on an improved bag-of-visual word model has the beneficial effects that the computation complexity is reduced; the space information between local features is kept; the image classification accuracy is improved; the classification time is reduced; and the method of the invention can be widely applied to fields such as image retrieval, object tracking, scene classification and behavior recognition.

Description

technical field [0001] The invention relates to the related technical field of computer graphic image classification technology, in particular to an image classification method based on an improved bag-of-words model. Background technique [0002] Image classification is one of the most challenging research tasks in computer vision. Image classification methods are of great significance to image retrieval, object tracking, scene classification and action recognition. The Bag of Visual Words model (Bag of Visual Words) originated from information retrieval and natural language processing. It regards text as an unordered collection of words, and uses the frequency information of words in the text to complete the classification of the text. Similar to text, images can be seen as a collection of local features that have nothing to do with location. These local features are equivalent to words in the text. In image classification, these words are called "visual words", and the co...

AI Technical Summary

Problems solved by technology

This method can well represent the texture structure of simple images, but it cannot guarantee the optimal visual dictionary for images with complex structures, such as scene images.

In addition, because the traditional BOVW visual dictionary cannot fully describe the global information of the image, the visual word histogram does not provide any spatial information, and the high dimensionality and sparsity of the feature space eventually lead to low classification accuracy

Images