Method and apparatus for detecting occluded image and medium

A device for detecting an occluded image, comprising:

• a processor;
• and
• a memory configured to store instructions executable by the processor;
• wherein the processor is configured to:
• after an image is captured by a camera, obtain the image as an image to be detected;
• input the image to be detected into a trained occluded-image detection model, wherein, the occluded-image detection model is trained based on original occluded images and non-occluded images by using a trained data feature augmentation network;
• determine whether the image to be detected is an occluded image based on the occluded-image detection model;
• and
• output an image detection result;
• wherein the processor is configured to:
• obtain the original occluded images and the non-occluded images;
• train the data feature augmentation network based on the original occluded images;
• and
• train the occluded-image detection model based on the original occluded images and the non-occluded images by using the trained data feature augmentation network;
• wherein the processor is configured to:
• generate a finger template image and a non-finger image based on each original occluded image, wherein the finger template image is obtained by removing an area not occluded by a finger from the original occluded image, and the non-finger image is obtained by removing an area occluded by the finger from the original occluded image;
• and
• generate a plurality of training data sets, wherein each training data set comprises one finger template image, one non-finger image and one original occluded image, and a plurality of training data sets are used to train the data feature augmentation network in each round of training;
• wherein the processor is configured to:
• obtain mask data based on the original occluded image, the mask data indicating an occluded location in the original occluded image;
• and
• generate the finger template image and the non-finger image based on the original occluded image and the mask data;
• wherein the processor is configured to:
• obtain a processed non-occluded image based on the mask data and the non-occluded image, wherein the processed non-occluded image is obtained by removing an area corresponding to the mask data from the non-occluded image;
• input the finger template image and the processed non-occluded image into the trained data feature augmentation network to generate a generation feature of the non-occluded image;
• input the original occluded image into a feature network to extract an original feature of the original occluded image, the occluded-image detection model comprising the feature network;
• input the generation feature of the non-occluded image and the original feature of the original occluded image into a classification network for training, the occluded-image detection model comprising the classification network;
• and
• determine that the training of the data feature augmentation network is finished when a loss function of the occluded-image detection model converges.

A non-transitory computer-readable storage medium having stored therein instructions that, when executed by a processor of a router, causes the router to perform a method for detecting an occluded image, the method comprising:

• after an image is captured by a camera, obtaining the image as an image to be detected;
• inputting the image to be detected into a trained occluded-image detection model, wherein, the occluded-image detection model is trained based on original occluded images and non-occluded images by using a trained data feature augmentation network;
• determining whether the image to be detected is an occluded image based on the occluded-image detection model;
• and
• outputting an image detection result;
• the method further comprising:
• obtaining the original occluded images and the non-occluded images;
• training the data feature augmentation network based on the original occluded images;
• and
• training the occluded-image detection model based on the original occluded images and the non-occluded images by using the trained data feature augmentation network;
• wherein training the data feature augmentation network based on the original occluded images comprises:
• generating a finger template image and a non-finger image based on each original occluded image, wherein the finger template image is obtained by removing an area not occluded by a finger from the original occluded image, and the non-finger image is obtained by removing an area occluded by the finger from the original occluded image;
• and
• generating a plurality of training data sets, wherein each training data set comprises one finger template image, one non-finger image and one original occluded image, and a plurality of training data sets are used to train the data feature augmentation network in each round of training;
• wherein generating the finger template image and the non-finger image based on each original occluded image comprises:
• obtaining mask data based on the original occluded image, the mask data indicating an occluded location in the original occluded image;
• and
• generating the finger template image and the non-finger image based on the original occluded image and the mask data;
• wherein training the occluded-image detection model based on the original occluded images and the non-occluded images by using the trained data feature augmentation network comprises:
• obtaining a processed non-occluded image based on the mask data and the non-occluded image, wherein the processed non-occluded image is obtained by removing an area corresponding to the mask data from the non-occluded image;
• inputting the finger template image and the processed non-occluded image into the trained data feature augmentation network to generate a generation feature of the non-occluded image;
• inputting the original occluded image into a feature network to extract an original feature of the original occluded image, the occluded-image detection model comprising the feature network;
• inputting the generation feature of the non-occluded image and the original feature of the original occluded image into a classification network for training, the occluded-image detection model comprising the classification network;
• and
• determining that the training of the data feature augmentation network is finished when a loss function of the occluded-image detection model converges.