2018/12/21 by Hichem Sahbi, Sahbi, Hichem
Computer Science · Earth and Planetary Sciences · Engineering · #Advanced Image and Video Retrieval Techniques #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Face and Expression Recognition #Image Retrieval and Classification Techniques #Remote Sensing and Land Use #Remote-Sensing Image Classification
paper · pdf · doi:10.48550/arxiv.1812.09280
openalex publication_date 2018/12/21 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Canonical correlation analysis (CCA) is a statistical learning method that\nseeks to build view-independent latent representations from multi-view data.\nThis method has been successfully applied to several pattern analysis tasks\nsuch as image-to-text mapping and view-invariant object/action recognition.\nHowever, this success is highly dependent on the quality of data pairing (i.e.,\nalignments) and mispairing adversely affects the generalization ability of the\nlearned CCA representations. In this paper, we address the issue of alignment\nerrors using a new variant of canonical correlation analysis referred to as\nalignment-agnostic (AA) CCA. Starting from erroneously paired data taken from\ndifferent views, this CCA finds transformation matrices by optimizing a\nconstrained maximization problem that mixes a data correlation term with\ncontext regularization; the particular design of these two terms mitigates the\neffect of alignment errors when learning the CCA transformations. Experiments\nconducted on multi-view tasks, including multi-temporal satellite image change\ndetection, show that our AA CCA method is highly effective and resilient to\nmispairing errors.\n