Facial Image De-Identification with Machine-Readable Identity Preservation

The patent applies different quality transformations to different parts of the facial image. Specifically, it modifies local facial features (eyes, nose, mouth, contours) with targeted geometric transformations while preserving overall facial structure and style characteristics. This allows machine recognition systems to maintain reliability by preserving essential facial patterns while making the image unrecognizable to humans through localized feature distortion.

The patent changes multiple parameters of facial features simultaneously including scaling factors, rotation angles, skew values, and translation vectors for different facial components. By adjusting these parameters within specific ranges and applying composite transformations, the system achieves de-identification that prevents human recognition while maintaining sufficient structural integrity for machine-based facial recognition algorithms to function.

The patent segments the facial image into distinct anatomical regions (eyes, eyebrows, nose, mouth, cheeks, jawline) and applies specific geometric transformations to each segment independently. This segmentation approach allows style transfer to effectively alter the overall appearance and prevent human recognition, while the systematic transformation of each segment preserves the underlying facial identity information that machine recognition systems can still process.

The patent extends traditional 2D image transformation by applying multi-dimensional geometric transformations including non-uniform scaling, rotation, skewing, and translation to facial features. These dimensional transformations alter the visual appearance significantly for human perception while maintaining the mathematical relationships and structural patterns that machine learning models rely upon for recognition.