Image recognition method and apparatus, device and non-volatile computer storage medium

A device, comprising

• one or more processor;
• a memory;
• one or more programs stored in the memory and configured to execute an image recognition method,
• wherein the method comprises:
• obtaining a to-be-recognized image of a designated size;
• extracting a different-area image from the to-be-recognized image;
• obtaining an image feature of the different-area image according to the different-area image;
• and
• obtaining a recognition result of the to-be-recognized image, according to the image feature of the different-area image and a preset template feature,
• wherein the extracting the different-area image from the to-be-recognized image comprises:
• positioning an area location from the to-be-recognized image according to a manually pre-designated area location having a larger differentiation, the to-be-recognized image containing a target object and other objects in addition to the target object, wherein the differentiation is relative to the other objects contained in the to-be-recognized image, and
• then extracting, only considering the area location of the to-be-recognized image containing only the target object, the image covered by the area location as the different-area image,
• wherein the extracting the different-area image from the to-be-recognized image further comprises employing a Scale-Invariant Feature Transform (SIFT) algorithm,
• wherein the obtaining the to-be-recognized image of the designated size comprises:
• using affine transform to adjust the obtained to-be-recognized image of any size as the to-be-recognized image of the designated size,
• wherein the using the affine transform comprises applying a series of atomic transformations to the obtained to-be-recognized image, the series of atomic transformations including translation, scale, flip, rotation, and shear,
• wherein the obtaining the image feature of the different-area image according to the different-area image comprises:
• obtaining the image feature of the different-area image by using a model obtained by training a Deep Neural Network from a general data set, according to the different-area image,
• wherein obtaining the recognition result of the to-be-recognized image comprises:
• obtaining template images of at least two designated classes;
• extracting a template area image of each designated class from the template image of said each designated class in the at least two designated classes according to the manually pre-designated area location;
• measuring a distance between the image feature of the different-area image and an image feature of each of the respective template area images;
• obtaining a template feature having a nearest distance among the measured distances;
• and
• regarding the designated class corresponding to the template feature having the nearest distance as the recognition result.

A non-transitory computer storage medium in which one or more programs are stored, an apparatus being enabled to execute an image recognition method, wherein the method comprises:

• obtaining a to-be-recognized image of a designated size;
• extracting a different-area image from the to-be-recognized image;
• obtaining an image feature of the different-area image according to the different-area image;
• and
• obtaining a recognition result of the to-be-recognized image, according to the image feature of the different-area image and a preset template feature,
• wherein the extracting the different-area image from the to-be-recognized image comprises:
• positioning an area location from the to-be-recognized image according to a manually pre-designated area location having a larger differentiation, the to-be-recognized image containing a target object and other objects in addition to the target object, wherein the differentiation is relative to other objects contained in the to-be-recognized image, and
• then extracting, only considering the area location of the to-be-recognized image containing only the target object, the image covered by the area location as the different-area image,
• wherein the extracting the different-area image from the to-be-recognized image further comprises employing a Scale-Invariant Feature Transform (SIFT) algorithm,
• wherein the obtaining the to-be-recognized image of the designated size comprises:
• using affine transform to adjust the obtained to-be-recognized image of any size as the to-be-recognized image of the designated size,
• wherein the using the affine transform comprises applying a series of atomic transformations to the obtained to-be-recognized image, the series of atomic transformations including translation, scale, flip, rotation, and shear,
• wherein the obtaining the image feature of the different-area image according to the different-area image comprises:
• obtaining the image feature of the different-area image by using a model obtained by training a Deep Neural Network from a general data set, according to the different-area image,
• wherein obtaining the recognition result of the to-be-recognized image comprises:
• obtaining template images of at least two designated classes;
• extracting a template area image of each designated class from the template image of said each designated class in the at least two designated classes according to the manually pre-designated area location;
• measuring a distance between the image feature of the different-area image and an image feature of each of the respective template area images;
• obtaining a template feature having a nearest distance among the measured distances;
• and
• regarding the designated class corresponding to the template feature having the nearest distance as the recognition result.