Training image recognition network, image recognition searching method and related device

Abstract

The invention discloses a training image recognition network, an image recognition searching method and a related device. The method for training the image recognition network comprises the followingsteps: segmenting an original training image into a plurality of training image blocks, and marking the training image blocks with labels; disrupting and rearranging the plurality of training image blocks according to an image salient region detection result of the original training image to obtain a rearranged training image of the original training image; and training an image recognition network to obtain an image recognition model by taking the original training image, the rearranged training image and corresponding annotation data including a coarse-grained image category label, a fine-grained image category label, an image preprocessing category label and a training image block label sequence as training data. The image recognition searching method comprises the steps of obtaining ato-be-recognized image; inputting a to-be-recognized image into the image recognition model, and outputting target features and target categories of the to-be-recognized image; and searching similar images in an image database by utilizing the target features and the target categories of the to-be-identified images.

Description

technical field [0001] The present application relates to the technical field of image processing, in particular to a method for training an image recognition network, image recognition search and related devices. Background technique [0002] With the rapid development of science and technology, in daily life, users can take pictures of their interested items, search for items with item images, and quickly obtain links to the same item or similar items, so as to meet the user's search needs for items of interest; among them , searching by item image is actually performing image recognition search on item image. [0003] At present, the above-mentioned image recognition and search methods generally use deep learning models to extract global features of item images for recognition and search. However, for an item image with a complex scene, for example, the item area in the item image is relatively small, the above-mentioned deep learning model can only extract the global fe...

AI Technical Summary

Problems solved by technology

However, for an item image with a relatively complex scene, for example, the item area in the item image is relatively small, the above-mentioned deep learning model can only extract the global features of the item image, and only focus on the item image in the subsequent image recognition search process. Global features, it is easy to miss the important features of the item image, which greatly reduces the accuracy of image recognition search, resulting in poor user experience of image recognition search

Images