Assessing the Potential for Trademarks to be Confused or Misrecognized by Using a Vision Transformer Algorithm to Evaluate the Similarity Calculation of Images

Assessing the Potential for Trademarks to be Confused or Misrecognized by Using a Vision Transformer Algorithm to Evaluate the Similarity Calculation of Images

This study proposes a method for assessing trademark confusion that measures trademark similarity using a misrecognition evaluation system based on the Vision Transformer (ViT) architecture. The proposed approach aims to enhance the objectivity and accuracy of trademark similarity evaluation, which has traditionally relied heavily on subjective human judgment and lacks unified standards. With the advancement of artificial intelligence, particularly in terms of image processing capabilities, ViT models provide a powerful tool for evaluating the visual similarity of novel graphic designs to existing trademarks. The proposed ViT model is based on a transformer architecture that captures both global and local features of images using self-attention mechanisms. Input images are divided into fixed-size patches that are processed through linear and positional embeddings and then passed through multiple transformer encoder layers. The resulting high-dimensional feature representations of the images are used to compute similarity scores between trademarks.
The proposed system first initializes a trained ViT model to extract features from input images. It then compares target trademark graphics with registered trademarks in a training dataset and ranks the results according to user-defined similarity thresholds. Experimental results demonstrate that the system successfully identified visually similar registered trademarks, displaying the top 200 candidates for each query. When the similarity threshold was set above 80%, the system exhibited high accuracy and practicality.
The contributions of this study are summarized as follows. First, the ViT model reduces the influence of individual judgement by learning statistical patterns from large-scale data. Second, it significantly increases the efficiency with which trademarks can be examined and thus reduces the human labor required to examine a given set of graphic designs for potential infringement. Third, the system can help trademark applicants in identifying potential infringement risks prior to registration, thereby minimizing legal disputes. Finally, the system provides computational and quantitative evidence to support claims of trademark confusion in legal proceedings. These results demonstrate the potential of artificial intelligence to enhance trademark examination and protection.