| CPC G06F 16/532 (2019.01) [G06F 16/55 (2019.01)] | 5 Claims |
|
1. An image retrieving method comprising the following steps:
a first step of generating a feature vector database for multiple target drawing images, where each image is processed by a first convolutional neural network (CNN) to create target feature vectors,
a second step of generating a query feature vector from a real-world photograph obtained by capturing a real object using a second convolutional neural network,
a third step of retrieving the target feature vector database for vectors most similar to the query feature vector generated from the query image,
the method including a generative model that creates a feature vector from a specific drawing image and a discriminative model that identifies differences between the feature vector extracted from the specific design drawing and those feature vectors extracted from real product images,
the generative model further iteratively adjusting the generation of the feature vector until the discriminative model finds it challenging to distinguish the generated feature vector from other feature vectors within the same cluster group,
the discriminative model undergoes adversarial training to maximize its ability to distinguish feature vectors extracted from multiple design drawing images from those feature vectors extracted from real product images of the items represented by the drawings,
the method further includes an additional feature vector generation step to create feature vectors for the same cluster group as a result of the adversarial training, enhancing the accuracy of similarity comparisons between drawing images and real product images.
|