G06F 18/21355 Full Title
Pattern recognition > Analysing > Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation > Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods > based on approximation criteria, e.g. principal component analysis
Description
G06F18/21355 covers nonlinear approximation criteria for feature extraction, in particular methods that embed a data manifold in a lower-dimensional Euclidean space. Typical techniques include kernel-based extensions of principal component analysis and manifold-learning approaches that recover a curved low-dimensional structure hidden in high-dimensional pattern data, used for visualisation, compression and preprocessing in recognition systems. Linear approximation methods such as classical PCA remain in the parent group G06F18/2135, and mappings driven by topology preservation are classified in G06F18/2137.
Filing statistics
CPC G06F 18/21355 belongs to the parent subclass G06F, which recorded 2,867,653 patent applications worldwide between 2013 and 2023. Annual filings grew from 154,451 in 2013 to 335,605 in 2023 (+117%), peaking at 367,101 applications in 2021. The most active applicants in class G06 are SAMSUNG ELECTRONICS COMPANY (76,952 applications), IBM (INTERNATIONAL BUSINESS MACHINES CORPORATION) (62,841 applications) and MICROSOFT TECHNOLOGY LICENSING (44,778 applications).
Source: EPO PATSTAT, worldwide filings
Top Applicants
Top 10 applicants by patent filingsfor class G06, 2013–2023, worldwide · Source: EPO PATSTAT
- SAMSUNG ELECTRONICS COMPANY KR 76,952
- IBM (INTERNATIONAL BUSINESS MACHINES CORPORATION) US 62,841
- MICROSOFT TECHNOLOGY LICENSING US 44,778
- GOOGLE US 35,735
- INTEL CORPORATION US 32,087
- HUAWEI TECHNOLOGIES COMPANY CN 30,572
- TENCENT TECHNOLOGY (SHENZHEN) COMPANY 25,023
- APPLE US 23,482
- SGCC(STATE GRID CORPORATION OF CHINA) 22,548
- HUAWEI TECHNOLOGIES COMPANY 20,917