Image labeling method based on convolutional neural network and binary coding features

Abstract

The invention discloses an image annotation method based on a convolutional neural network and binary coding features. The method comprises the following steps: constructing an Incepton V3 basic network model; intercepting a final pooling layer of the Incepton V3 network basic model; removing Logits and softmax functions of the Incepon V3 network basic model, and using a sigmoid function as an activation function of the last layer to obtain a modified first basic network model; adding two full connection layers on the first basic network model, and using a sigmoid function as an activation function of the last layer to obtain a multi-label classification network model; performing training learning on the training set by using a multi-label classification network model, and optimizing the weight of the multi-label classification network model; marking the feature vector set of the target image based on the trained multi-label classification network model to obtain multi-label probability output of the target image; and in combination with multi-label probability output, labeling the target image by adopting a TagProp algorithm. Multi-label labeling of images can be realized, the cost is low, and the efficiency is high.

Description

technical field [0001] The invention relates to the technical field of visual images, in particular to an image labeling method based on convolutional neural network and binary coded features. Background technique [0002] In order to realize the effective management and retrieval of large-scale images, the efficient annotation of images is becoming more and more important. The goal of image annotation is to assign a set of relevant descriptive labels to images. Traditional image annotation algorithms need to spend a lot of time manually extracting image features, and may not achieve good results, so deep learning is applied to image annotation. Deep learning can obtain higher-level semantic features of images, which narrows the difference with high-level semantic concepts such as labels. The automatic image labeling algorithm based on deep learning does not need to manually extract image features, so that the labeling algorithm is no longer subject to the choice of featur...

AI Technical Summary

Problems solved by technology

In actual situations, an image is often associated with multiple tags, and a single tag cannot fully describe the entire image.

The current convolutional neural network models are also constructed based on single-label image classification tasks. The loss function is usually a Softmax function, and only the label with the highest probability can be assigned to the image.

Images