A two-part module project applying classic signal and image processing techniques to a running theme of Japan-related datasets — from Tokyo weather readings to seismic traces, Go boards, and torii gates.
Mini-Challenge 1 — Signals covers the foundations: the Nyquist–Shannon sampling theorem is explored using hourly Tokyo temperature data, testing how far the series can be safely downsampled before the daily cycle breaks down. Correlation is examined on the same weather data, and convolution-based filtering is applied to real seismic acceleration traces recorded in Nagano, Japan.
Mini-Challenge 2 — Images moves into the spatial domain: a full pipeline segments and counts Go stones from overhead board photos using K-means segmentation, morphological operations (including distance-transform-based erosion), and object property extraction. A separate notebook explores image augmentation techniques (CLAHE, FFT-based high-boost filtering, Reinhard colour transfer) on scenes like Shibuya Crossing and cherry blossoms, while a pattern detection notebook applies edge detection to identify torii gates.
Both mini-challenges close with a reflection on what worked, what proved trickier than expected, and how the choices behind small preprocessing steps (zero-padding, border handling, kernel size) can noticeably change downstream results.