Research And Design On Multiple-objective Image Annotation Base On Neural Network

Nowadays,text search is getting more and more mature,the requirements for image search are also growing.The image index and retrieval are very difficult because the image has more information than the text and the storage method of image is difficult to reflect the semantic features.Using mature text search,there is a good idea that image search is changed into text search.But this method is dependent on the exact description of the image semantics.This description grows rapidly with the number of images.Manual methods take a long time and are expensive.It can't meet people's requirements.In order to make a better image retrieval technology,automatic image captioning method is very important and urgent.At the same time,the automatic image captioning algorithm has important applications in many fields.For example,video search and the movie plot description for the blind,and so on.In this paper,the method of automatic image captioning is researched using neural network.The details are listed below:1.The automatic image captioning is dependent on the object recognition.Only recognizing the object in the image,we generate the annotation information.And the objects in an image may be a lot,each object has a lot of attributes,even some objects contain sub-objects.We must sample in a large number of objects,select the important object and start recognition object.2.We will introduce the automatic image captioning algorithm,which requires a computer vision system to both object recognition and natural language generation.The object recognition model is using the CNN model.While the natural language model is using the RNN model.A complete system is formed with them together.