Examples with BlockAverageDataProcessor gdsc.smlm.filters.BlockAverageDataProcessor used on opensource projects

Example 1 with BlockAverageDataProcessor

use of gdsc.smlm.filters.BlockAverageDataProcessor in project GDSC-SMLM by aherbert.

the class FitWorker method run.

/**
	 * Locate all the peaks in the image specified by the fit job
	 * <p>
	 * WARNING: The FitWorker fits a sub-region of the data for each maxima. It then updates the FitResult parameters
	 * with an offset reflecting the position. The initialParameters are not updated with this offset unless configured.
	 * 
	 * @param job
	 *            The fit job
	 */
public void run(FitJob job) {
    final long start = System.nanoTime();
    job.start();
    this.job = job;
    benchmarking = false;
    this.slice = job.slice;
    // Used for debugging
    //if (logger == null) logger = new gdsc.fitting.logging.ConsoleLogger();
    // Crop to the ROI
    cc = new CoordinateConverter(job.bounds);
    final int width = cc.dataBounds.width;
    final int height = cc.dataBounds.height;
    borderLimitX = width - border;
    borderLimitY = height - border;
    data = job.data;
    FitParameters params = job.getFitParameters();
    this.endT = (params != null) ? params.endT : -1;
    candidates = indentifySpots(job, width, height, params);
    if (candidates.size == 0) {
        finish(job, start);
        return;
    }
    fittedBackground = 0;
    // Always get the noise and store it with the results.
    if (params != null && !Float.isNaN(params.noise)) {
        noise = params.noise;
        fitConfig.setNoise(noise);
    } else if (calculateNoise) {
        noise = estimateNoise(width, height);
        fitConfig.setNoise(noise);
    }
    //System.out.printf("Slice %d : Noise = %g\n", slice, noise);
    if (logger != null)
        logger.info("Slice %d: Noise = %f", slice, noise);
    final ImageExtractor ie = new ImageExtractor(data, width, height);
    double[] region = null;
    if (params != null && params.fitTask == FitTask.MAXIMA_IDENITIFICATION) {
        final float sd0 = (float) fitConfig.getInitialPeakStdDev0();
        final float sd1 = (float) fitConfig.getInitialPeakStdDev1();
        for (int n = 0; n < candidates.getSize(); n++) {
            // Find the background using the perimeter of the data.
            // TODO - Perhaps the Gaussian Fitter should be used to produce the initial estimates but no actual fit done.
            // This would produce coords using the centre-of-mass.
            final Candidate candidate = candidates.get(n);
            final int x = candidate.x;
            final int y = candidate.y;
            final Rectangle regionBounds = ie.getBoxRegionBounds(x, y, fitting);
            region = ie.crop(regionBounds, region);
            final float b = (float) Gaussian2DFitter.getBackground(region, regionBounds.width, regionBounds.height, 1);
            // Offset the coords to the centre of the pixel. Note the bounds will be added later.
            // Subtract the background to get the amplitude estimate then convert to signal.
            final float amplitude = candidate.intensity - ((relativeIntensity) ? 0 : b);
            final float signal = (float) (amplitude * 2.0 * Math.PI * sd0 * sd1);
            final float[] peakParams = new float[] { b, signal, 0, x + 0.5f, y + 0.5f, sd0, sd1 };
            final int index = y * width + x;
            sliceResults.add(createResult(cc.fromDataToGlobalX(x), cc.fromDataToGlobalY(y), data[index], 0, noise, peakParams, null, n));
        }
    } else {
        initialiseFitting();
        // Smooth the data to provide initial background estimates
        final BlockAverageDataProcessor processor = new BlockAverageDataProcessor(1, 1);
        final float[] smoothedData = processor.process(data, width, height);
        final ImageExtractor ie2 = new ImageExtractor(smoothedData, width, height);
        // Allow the results to be filtered for certain peaks
        if (params != null && params.filter != null) {
            resultFilter = new DistanceResultFilter(params.filter, params.distanceThreshold, candidates.getlength());
        //filter = new OptimumDistanceResultFilter(params.filter, params.distanceThreshold, maxIndices.length);
        }
        // The SpotFitter is used to create a dynamic MultiPathFitResult object.
        // This is then passed to a multi-path filter. Thus the same fitting decision process 
        // is used when benchmarking and when running on actual data.
        // Note: The SpotFitter labels each PreprocessedFitResult using the offset in the FitResult object.
        // The initial params and deviations can then be extracted for the results that pass the filter.
        MultiPathFilter filter;
        IMultiPathFitResults multiPathResults = this;
        SelectedResultStore store = this;
        coordinateStore = coordinateStore.resize(width, height);
        if (params != null && params.fitTask == FitTask.BENCHMARKING) {
            // Run filtering as normal. However in the event that a candidate is missed or some
            // results are not generated we must generate them. This is done in the complete(int)
            // method if we set the benchmarking flag.
            benchmarking = true;
            // Filter using the benchmark filter
            filter = params.benchmarkFilter;
            if (filter == null) {
                // Create a default filter using the standard FitConfiguration to ensure sensible fits
                // are stored as the current slice results.
                // Note the current fit configuration for benchmarking may have minimal filtering settings
                // so we do not use that object.
                final FitConfiguration tmp = new FitConfiguration();
                final double residualsThreshold = 0.4;
                filter = new MultiPathFilter(tmp, createMinimalFilter(), residualsThreshold);
            }
        } else {
            // Filter using the configuration
            filter = new MultiPathFilter(fitConfig, createMinimalFilter(), config.getResidualsThreshold());
        }
        // If we are benchmarking then do not generate results dynamically since we will store all 
        // results in the fit job
        dynamicMultiPathFitResult = new DynamicMultiPathFitResult(ie, ie2, !benchmarking);
        // Debug where the fit config may be different between benchmarking and fitting
        if (slice == -1) {
            SettingsManager.saveFitEngineConfiguration(config, String.format("/tmp/config.%b.xml", benchmarking));
            Utils.write(String.format("/tmp/filter.%b.xml", benchmarking), filter.toXML());
            //filter.setDebugFile(String.format("/tmp/fitWorker.%b.txt", benchmarking));
            StringBuilder sb = new StringBuilder();
            sb.append((benchmarking) ? ((gdsc.smlm.results.filter.Filter) filter.getFilter()).toXML() : fitConfig.getSmartFilter().toXML()).append("\n");
            sb.append(((gdsc.smlm.results.filter.Filter) filter.getMinimalFilter()).toXML()).append("\n");
            sb.append(filter.residualsThreshold).append("\n");
            sb.append(config.getFailuresLimit()).append("\n");
            sb.append(fitConfig.getDuplicateDistance()).append("\n");
            if (spotFilter != null)
                sb.append(spotFilter.getDescription()).append("\n");
            sb.append("MaxCandidate = ").append(candidates.getSize()).append("\n");
            for (int i = 0; i < candidates.list.length; i++) {
                sb.append(String.format("Fit %d [%d,%d = %.1f]\n", i, candidates.get(i).x, candidates.get(i).y, candidates.get(i).intensity));
            }
            Utils.write(String.format("/tmp/candidates.%b.xml", benchmarking), sb.toString());
        }
        filter.select(multiPathResults, config.getFailuresLimit(), true, store, coordinateStore);
        if (logger != null)
            logger.info("Slice %d: %d / %d", slice, success, candidates.getSize());
        // Result filter post-processing
        if (resultFilter != null) {
            resultFilter.finalise();
            job.setResults(resultFilter.getResults());
            job.setIndices(resultFilter.getMaxIndices());
            for (int i = 0; i < resultFilter.getFilteredCount(); i++) {
                job.setFitResult(i, resultFilter.getFitResults()[i]);
            }
            sliceResults.clear();
            sliceResults.addAll(resultFilter.getResults());
        }
    }
    // Add the ROI bounds to the fitted peaks
    final float offsetx = cc.dataBounds.x;
    final float offsety = cc.dataBounds.y;
    for (int i = 0; i < sliceResults.size(); i++) {
        final PeakResult result = sliceResults.get(i);
        result.params[Gaussian2DFunction.X_POSITION] += offsetx;
        result.params[Gaussian2DFunction.Y_POSITION] += offsety;
    }
    this.results.addAll(sliceResults);
    finish(job, start);
}