###### tags: `jptw` `thesis` `meeting` # 20200812 DSP Group Meeting ## Review Work ### Fingerprinting 1. Find Peaks in spectrogram Pass through a HP Filter and then find peaks with ["Dilation and Erosion"]((https://sites.google.com/a/ms.ttu.edu.tw/cse2012dance-robot/yan-jiu-cheng-guo/opencv-ruan-ti-she-ji/qin-shi-yu-peng-zhang)). ![](https://i.imgur.com/QYxljS3.png =49%x) ![](https://i.imgur.com/dKDD3lU.png =49%x) ![](https://i.imgur.com/fMLOIlo.png =49%x) ![](https://i.imgur.com/YJBb9xo.png =49%x) 2. Record each peak with the temporal difference among its neighborhood, and hash with 'SHA1' algorithm. ::: warning The beauty is that a good hash function will not only return the same output integer each time the input is the same, but also that very few different inputs will have the same output. ::: **TODO** > Add time information for fingerprinting in DB > Traditional: Clear source, dirty target > Our case: **Dirty** source, clear target ## Feedback - Good point for preprocessing with noise reduction!