Local image feature and multi-instance learning-based family photo and non family photo classification method

Abstract

The invention discloses a local image feature and multi-instance learning-based family photo and non family photo classification method. The method comprises the following steps of S1: extracting local features of each photo, wherein the local features are local geometric features, local family relation features or local semantic features; S2: selecting a multi-instance learning framework-based binary classifier, and by taking all the local features and a photo tag of each photo as inputs, training parameters of the multi-instance classifier; and S3: in a test stage, inputting local features of photos of unknown types to the trained multi-instance classifier in the step S2, thereby obtaining predicted types. According to the method, discriminative information in local regions of the photos can be effectively utilized; the information is ensured not to be polluted by useless information; and the classification precision is high.

Description

technical field [0001] The invention relates to the technical fields of pattern recognition, computer vision and multimedia analysis, in particular to a method for classifying family photos and non-family photos based on local image features and multi-instance learning. Background technique [0002] In the existing image-based classification algorithms for family photos and non-family photos, all methods use global feature representation and classification for photo recognition. Specifically, the existing method first extracts a global feature vector from each photo, summarizes all the discriminative information in the photo into the vector, and then uses a traditional binary classifier for training and photo recognition. However, the most discriminative information in family photos and non-family photos is often located in the local part of the image, and the global feature representation adopted by existing methods will mix discriminative information and useless informatio...

AI Technical Summary

Problems solved by technology

However, the most discriminative information in family photos and non-family photos is often located in the local part of the image, and the global feature representation adopted by existing methods will mix discriminative information and useless information, weakening the role of useful information and affecting classification accuracy

Therefore, there is a big deficiency in the above method

Images