Technical articles and historical engineering writing on software and systems engineering, real-time data processing, artificial intelligence, speech and signal processing, digital forensics, cybersecurity and algorithms.
The Blog contains my technical articles on software and systems engineering, artificial intelligence, digital forensics, cybersecurity, and the history of technology.
A multimodal view of lip reading: temporal visual speech representation, viseme ambiguity and fusion with the acoustic channel.
In a lightweight reflection-based ORM, the important costs come from metadata mapping, type conversion, transaction boundaries, and reuse of reflection metadata rather than SQL string generation alone.
Face and person detection with alignment, biometric enrollment, verification and identification, embedding-based matching, open-set decisions, thresholds, and error rates.
Depth and distance estimation through stereo geometry, disparity, monocular cues and sensor fusion, including the sources of metric uncertainty.
Object detection engineering across one- and two-stage detectors, IoU, NMS, boundary extraction and end-to-end latency.
AbstractRoutingDataSource does not create a pool; it routes a connection request to one of the independent HikariCP pools according to the key selected before the transaction starts.
Pattern recognition and OCR from visual representations to language-aware decoding, with character- and word-level error analysis.
Visual anomaly detection through models of normality, anomaly scores, temporal context, false-alarm control and distribution shift.
A technical framework for resampling, generative models, augmentation and identity transfer, with a clear distinction between plausible synthesis and evidential reconstruction.
Image segmentation from thresholding and morphology to learned encoder-decoder models, with emphasis on mask quality, annotation noise and runtime cost.
The technical boundary between image restoration and enhancement across denoising, deblurring, contrast processing, inpainting and super-resolution.
A keyboard-wedge RFID reader usually delivers keystrokes rather than an RFID protocol frame, so .NET UID capture must handle timing, focus, keyboard layout, and device identity explicitly.
Uses only this site's published content. How it works
