Particle picking Carlos Óscar S. Sorzano Vahid Abrishami Instruct Image Processing Center
Particle picking The problem Preprocessing Automatic picking – 3D Model-based picking – 2D Model-based picking – Feature-based picking Screening Consensus picking
The problem
Preprocessing Downsampling Fourier filtering Wavelet filtering Quantization
Automatic picking: 3D model based Correlation peaks: Cross-correlation Fourier-correlation Local-correlation Normalized-correlatio n Threshold criteria:
Automatic picking: 2D model based Correlation peaks: Cross-correlation Fourier-correlation Local-correlation Normalized-correlatio n Threshold criteria:
Automatic picking: Feature based 91D vector
Automatic picking: Feature based Classifier: SVM Naive Bayesian Neural network LDA Cascaded classifiers: AdaBoost
Manual supervision
Automatic Screening 20D vector
Screening: Mahalanobis distance
Automatic Screening
Consensus picking
Conclusions Picking families: – 2D/3D Model based: “correlation”+threshold criterion – Feature based: nD features+classifier A posteriori screening: – nD features+distance rank Consensus picking State-of-the-art: 85% precision, 70% recall