Method and system of classifying similar images

Abstract

The invention provides a method and a system of classifying similar images. The method comprises the following steps: inputting images to be identified and obtaining shape features, gradient features, color features and texture features of the images to be identified; segmenting a training sample in an image base into a plurality of local area images different in size, and conducting size conversion so as to obtain an image template set which comprises a plurality of image templates; analyzing the image templates in the image template set and obtaining shape features, gradient features, color features and texture features of the image templates; matching and processing corresponding features of the images to be identified and the images in the image template set so as to obtain image detailed information of the images to be identified; obtaining classifications of the images to be identified through image presenting data and by utilization of a bagging classifier. Accurate classification of similar images is realized through extracting and matching the shape features, the gradient features, the color features and the texture features of the input images according to the method.

Description

technical field [0001] The invention relates to the technical field of image recognition, in particular to a similar image classification method and system. Background technique [0002] Classification is essentially a cascading form of a tree structure. In this structure, a higher-level node near the root describes an inclusive class, also known as a super-ordinate class, such as a vehicle. The middle layer, also called the basic class (basicclass) node describes more specific categories, for example, motorcycles or cars. The lower-level nodes near the leaf nodes, also called subordinate classes, usually capture more subtle differences between objects, such as sports motorcycles or utility motorcycles, passenger cars, trucks, or cars. Similar image classification refers to classifying objects of the same basic class or aspects such as shape and visual appearance and their similarity, that is, classifying objects of subordinate categories, for example, distinguishing differ...

AI Technical Summary

Problems solved by technology

Although the "terms" obtained by clustering using k-means or Gaussian mixture model sometimes appear with high probability in a certain spatial area, they are not necessarily useful information for category judgment.

Second, when the detected region is mapped into a "dictionary entry" form, a lot of detailed information is lost

Third, the "dictionary" method needs to manually adjust some parameters of the clustering, which is cumbersome and may not be particularly suitable

These methods are effective for identifying attributes like fur, points, or four legs, but are less than ideal for distinguishing subtle differences between subordinate layer objects

Images