WiMi Hologram Cloud Inc, a leading global Hologram Augmented Reality (“AR”) Technology provider, announced that it developed a Mask R-CNN-based technique for intelligently extracting CSOs (feature space objects) and its reference points brings a breakthrough in the field of high-resolution image processing and matching. The technique utilizes the latest advances in deep learning and computer vision to provide an efficient and accurate solution for automatic image matching and target localization.
High-resolution image processing and matching have been an important research direction in the field of computer vision, but automatic matching has been facing great challenges due to local deformations in images and differences in lighting conditions. Previous methods are often limited by computational complexity and dependence on local features, making it difficult to achieve accurate results. WiMi’s technique can be used to extract CSOs and their reference points on images. With this method, the CSOs can be acquired automatically and provide accurate localization information for the subsequent image matching process.
Also Read: TechSee Brings Computer Vision AI and Augmented Reality Service Automation to Amazon Connect
WiMi’s R&D team successfully solved this challenge by introducing the Mask R-CNN model, a model extension based on Faster R-CNN commonly used for target detection and instance segmentation. The model is unique in that it can simultaneously predict the bounding box, category, mask and key points of a target, providing comprehensive information for image processing tasks.
In this new technique, WiMi first utilizes a large amount of high-resolution remote sensing image data for training the Mask R-CNN model. Through training, the model is able to learn the features of different target instances in the image and accurately predict their bounding boxes, categories, masks and key points. Based on the trained Mask R-CNN model, the technical team further proposes the concept of CSO and the reference point method. CSO refers to target instances with distinctive features, which can be intelligently filtered out by setting thresholds or rules. Reference points, on the other hand, are extracted from CSOs by a mask predictor and a key point predictor, which are used to locate important feature points of target instances.
SOURCE: PRNewswire