def runner(parser, options, args): if not hasattr(parser, 'runner'): options.output_path = None if args: raise NotImplementedError( ` args `) if len(args) == 1: if options.input_path: print >> sys.stderr, "WARNING: overwriting input path %r with %r" % ( options.input_path, args[0]) options.input_path = args[0] elif len(args) == 2: if options.input_path: print >> sys.stderr, "WARNING: overwriting input path %r with %r" % ( options.input_path, args[0]) options.input_path = args[0] options.output_path = args[1] else: parser.error( "incorrect number of arguments (expected upto 2 but got %s)" % (len(args))) options.kernel_path = fix_path(options.kernel_path) options.input_path = fix_path(options.input_path) if options.output_path is None: b, e = os.path.splitext(options.input_path) suffix = '_convolved' % () options.output_path = b + suffix + (e or '.tif') options.output_path = fix_path(options.output_path) kernel = ImageStack.load(options.kernel_path, options=options) stack = ImageStack.load(options.input_path, options=options) result = convolve(kernel.images, stack.images, options=options) if 1: print 'Saving result to', options.output_path ImageStack(result, stack.pathinfo).save(options.output_path)
def runner(parser, options, args): verbose = options.verbose if options.verbose is not None else True if not hasattr(parser, 'runner'): options.output_path = None if args: if len(args) == 1: if options.input_path: print >> sys.stderr, "WARNING: overwriting input path %r with %r" % ( options.input_path, args[0]) options.input_path = args[0] elif len(args) == 2: if options.input_path: print >> sys.stderr, "WARNING: overwriting input path %r with %r" % ( options.input_path, args[0]) options.input_path = args[0] if options.output_path: print >> sys.stderr, "WARNING: overwriting output path %r with %r" % ( options.output_path, args[1]) options.output_path = args[1] else: parser.error( "incorrect number of arguments (expected upto 2 but got %s)" % (len(args))) options.input_path = fix_path(options.input_path) kernel_width = options.kernel_width or None kernel_type = options.kernel smoothing_method = options.method boundary_condition = options.boundary stack = ImageStack.load(options.input_path, options=options) voxel_sizes = stack.get_voxel_sizes() if kernel_width is None: dr = stack.get_lateral_resolution() dz = stack.get_axial_resolution() if dr is None or dz is None: kernel_width = 3 else: print 'lateral resolution: %.3f um (%.1f x %.1f px^2)' % ( 1e6 * dr, dr / voxel_sizes[1], dr / voxel_sizes[2]) print 'axial resolution: %.3f um (%.1fpx)' % (1e6 * dz, dz / voxel_sizes[0]) vz, vy, vx = voxel_sizes m = 1 scales = (m * vz / dz, m * vy / dr, m * vx / dr) if kernel_width is not None: w = float(kernel_width) * min(voxel_sizes) scales = tuple([s / w for s in voxel_sizes]) print 'Window sizes:', [1 / s for s in scales] kdims = [1 + 2 * (int(numpy.ceil(1 / s)) // 2) for s in scales] k = 'x'.join(map(str, kdims)) if options.output_path is None: b, e = os.path.splitext(options.input_path) suffix = '_%s_%s%s' % (smoothing_method, kernel_type, k) options.output_path = b + suffix + (e or '.tif') options.output_path = fix_path(options.output_path) if options.link_function == 'identity': images = stack.images elif options.link_function == 'log': images = stack.images mn, mx = get_dtype_min_max(images.dtype) images = numpy.log(images) images[numpy.where(numpy.isnan(images))] = numpy.log(mn) else: raise NotImplementedError( ` options.link_function `) new_images, new_images_grad = regress(images, scales, kernel=kernel_type, method=smoothing_method, boundary=boundary_condition, verbose=verbose) if options.link_function == 'identity': pass elif options.link_function == 'log': new_images = numpy.exp(new_images) new_images = numpy.nan_to_num(new_images) else: raise NotImplementedError( ` options.link_function `) if verbose: print 'Leak: %.3f%%' % (100 * (1 - new_images.sum() / stack.images.sum())) print 'MSE:', ((new_images - stack.images)**2).mean() print 'Energy:', ((stack.images)**2).sum() print 'Saving result to', options.output_path ImageStack(new_images, stack.pathinfo).save(options.output_path)
def runner(parser, options, args): if not hasattr(parser, 'runner'): options.output_path = None if args: if len(args) == 1: if options.input_path: print >> sys.stderr, "WARNING: overwriting input path %r with %r" % ( options.input_path, args[0]) options.input_path = args[0] elif len(args) == 2: if options.input_path: print >> sys.stderr, "WARNING: overwriting input path %r with %r" % ( options.input_path, args[0]) options.input_path = args[0] if options.output_path: print >> sys.stderr, "WARNING: overwriting output path %r with %r" % ( options.output_path, args[1]) options.output_path = args[1] else: parser.error( "Incorrect number of arguments (expected upto 2 but got %s)" % (len(args))) if options.input_path is None: parser.error('Expected --input-path but got nothing') options.input_path = fix_path(options.input_path) stack = ImageStack.load(options.input_path, options=options) numpy_types = numpy.typeDict.values() if options.output_type in ['<detect>', None]: if str(stack.images.dtype).startswith('float'): output_type_name = 'float32' elif str(stack.images.dtype).startswith('int'): output_type_name = 'int32' elif str(stack.images.dtype).startswith('uint'): output_type_name = 'uint32' else: output_type_name = 'int32' else: output_type_name = options.output_type.lower() output_type = getattr(numpy, output_type_name, None) mn, mx = stack.images.min(), stack.images.max() print 'Input minimum and maximum: %s, %s' % (mn, mx) if options.scale and 'int' in output_type_name: tmn, tmx = get_dtype_min_max(output_type) new_images = (tmn + float(tmx - tmn) * (stack.images - float(mn)) / (mx - mn)).astype(output_type) else: new_images = stack.images.astype(output_type) print 'Output minimum and maximum: %s, %s' % (new_images.min(), new_images.max()) output_path = options.output_path output_ext = options.output_ext if output_path is None: dn = os.path.dirname(options.input_path) bn = os.path.basename(options.input_path) if os.path.isfile(options.input_path): fn, ext = os.path.splitext(bn) type_part = None for t in numpy_types: if fn.endswith('_' + t.__name__): type_part = t.__name__ break if type_part is None: output_path = os.path.join( dn, fn + '_' + output_type_name + '.' + output_ext) else: output_path = os.path.join( dn, fn[:-len(type_part)] + output_type_name + '.' + output_ext) elif os.path.isdir(options.input_path): output_path = os.path.join( dn, bn + '_' + output_type_name + '.' + output_ext) else: raise NotImplementedError('%s is not file nor directory' % (options.input_path)) output_path = fix_path(output_path) print 'Saving new stack to', output_path if output_ext == 'tif': ImageStack(new_images, stack.pathinfo, options=options).save(output_path) elif output_ext == 'data': from iocbio.microscope.psf import normalize_unit_volume, discretize value_resolution = stack.pathinfo.get_value_resolution() normal_images = normalize_unit_volume(new_images, stack.get_voxel_sizes()) discrete = discretize(new_images / value_resolution) signal_indices = numpy.where(discrete > 0) new_value_resolution = value_resolution * normal_images.max( ) / new_images.max() ImageStack(normal_images, stack.pathinfo, value_resolution=new_value_resolution).save( output_path, zip(*signal_indices)) elif output_ext == 'vtk': from pyvtk import VtkData, StructuredPoints, PointData, Scalars vtk = VtkData(StructuredPoints(new_images.shape), PointData(Scalars(new_images.T.ravel()))) vtk.tofile(output_path, 'binary') else: raise NotImplementedError( ` output_ext `)
import numpy from iocbio.ops import convolve from iocbio.microscope.deconvolution import deconvolve from iocbio.io import ImageStack import scipy.stats from matplotlib import pyplot as plt kernel = numpy.array([0, 1, 3, 1, 0]) test_data = numpy.array([0, 0, 0, 0, 2, 0, 0, 0, 0, 0, 1, 0, 1, 0, 0, 0]) * 50 data = convolve(kernel, test_data) degraded_data = scipy.stats.poisson.rvs(numpy.where(data <= 0, 1e-16, data)).astype(data.dtype) psf = ImageStack(kernel, voxel_sizes=(1, )) stack = ImageStack(degraded_data, voxel_sizes=(1, )) deconvolved_data = deconvolve(psf, stack).images plt.plot(test_data, label='test') plt.plot(data, label='convolved') plt.plot(degraded_data, label='degraded') plt.plot(deconvolved_data, label='deconvolved') plt.legend() plt.ylabel('data') plt.xlabel('index') plt.title('Deconvolving degraded test data.') plt.savefig('deconvolve_poisson_1d.png') plt.show()
def runner(parser, options, args): if not hasattr(parser, 'runner'): options.output_path = None options.input_path = fix_path(options.input_path) stack = ImageStack.load(options.input_path, options=options) voxel_sizes = stack.get_voxel_sizes() dr = stack.get_lateral_resolution() dz = stack.get_axial_resolution() if dr is not None: print 'lateral resolution: %.3f um (%.1f x %.1f px^2)' % ( 1e6 * dr, dr / voxel_sizes[1], dr / voxel_sizes[2]) print 'axial resolution: %.3f um (%.1fpx)' % (1e6 * dz, dz / voxel_sizes[0]) vz, vy, vx = voxel_sizes m = 1 scales = (m * vz / dz, m * vy / dr, m * vx / dr) else: raise NotImplementedError('get_lateral_resolution') kdims = [1 + 2 * (int(numpy.ceil(1 / s)) // 2) for s in scales] k = 'x'.join(map(str, kdims)) print 'Averaging window box:', k kernel_type = options.kernel smoothing_method = options.method boundary_condition = options.boundary mn, mx = stack.images.min(), stack.images.max() high_indices = numpy.where(stack.images >= mn + 0.9 * (mx - mn)) high = stack.images[high_indices] from iocbio.ops import regress average, average_grad = regress(stack.images, scales, kernel=kernel_type, method=smoothing_method, boundary=boundary_condition, verbose=True, enable_fft=True) ImageStack(average, pathinfo=stack.pathinfo).save('average.tif') noise = stack.images - average ImageStack(noise - noise.min(), pathinfo=stack.pathinfo).save('noise.tif') bright_level = 0.999 * average.max() + 0.001 * average.min() bright_indices = numpy.where(average >= bright_level) print len(bright_indices[0]) bright_noise = stack.images[bright_indices] - average[bright_indices] a = stack.images[bright_indices].mean() d = stack.images[bright_indices].std() print 'mean=', a, 'std=', d print 'peak SNR=', a / d print 'AVERAGE min, max, mean = %s, %s, %s' % ( average.min(), average.max(), average.mean()) print numpy.histogram(stack.images)[0] sys.exit() noise = stack.images - average var, var_grad = regress(noise * noise, scales, kernel=kernel_type, method=smoothing_method, boundary=boundary_condition, verbose=True, enable_fft=True) print 'VAR min, max, mean = %s, %s, %s' % (var.min(), var.max(), var.mean()) indices = numpy.where(var > 0) print len(numpy.where(var == 0)[0]), var.shape, var.dtype var[numpy.where(var <= 0)] = 1 snr = average / numpy.sqrt(var) snr1 = snr[indices] print 'STACK min, max = %s, %s' % (mn, mx) print 'SNR min, max, mean = %s, %s, %s' % (snr1.min(), snr1.max(), snr1.mean()) ImageStack(average, pathinfo=stack.pathinfo).save('average.tif') ImageStack(snr, pathinfo=stack.pathinfo).save('snr.tif') ImageStack(noise - noise.min(), pathinfo=stack.pathinfo).save('noise.tif')
def runner (parser, options, args): if not hasattr(parser, 'runner'): options.output_path = None if args: if len (args)==1: if options.input_path: print >> sys.stderr, "WARNING: overwriting input path %r with %r" % (options.input_path, args[0]) options.input_path = args[0] elif len(args)==2: if options.input_path: print >> sys.stderr, "WARNING: overwriting input path %r with %r" % (options.input_path, args[0]) options.input_path = args[0] if options.output_path: print >> sys.stderr, "WARNING: overwriting output path %r with %r" % (options.output_path, args[1]) options.output_path = args[1] else: parser.error("Incorrect number of arguments (expected upto 2 but got %s)" % (len(args))) if options.input_path is None: parser.error('Expected --input-path but got nothing') options.input_path = fix_path (options.input_path) stack = ImageStack.load(options.input_path, options=options) numpy_types = numpy.typeDict.values() if options.output_type in ['<detect>', None]: output_type_name = stack.images.dtype.name else: output_type_name = options.output_type.lower() output_type = getattr (numpy, output_type_name, None) nof_stacks = stack.get_nof_stacks() old_shape = stack.images.shape new_shape = (nof_stacks, old_shape[0]//nof_stacks) + old_shape[1:] new_images = numpy.zeros (new_shape[1:], dtype=output_type_name) first_stack = None last_stack = None for i, stacki in enumerate(stack.images.reshape(new_shape)): if i==0: first_stack = stacki.astype (float) new_images[:] = stacki else: err_first = abs(stacki - first_stack).mean() err_last = abs(stacki - last_stack).mean() print ('Stack %i: mean abs difference from first and last stack: %.3f, %.3f' % (i+1, err_first, err_last)) new_images += stacki last_stack = stacki.astype(float) output_path = options.output_path output_ext = options.output_ext if output_path is None: dn = os.path.dirname(options.input_path) bn = os.path.basename(options.input_path) if os.path.isfile(options.input_path): fn, ext = os.path.splitext (bn) fn += '_sumstacks%s' % (nof_stacks) type_part = None for t in numpy_types: if fn.endswith('_' + t.__name__): type_part = t.__name__ break if type_part is None: output_path = os.path.join(dn, fn + '_' + output_type_name + '.' + output_ext) else: output_path = os.path.join(dn, fn[:-len(type_part)] + output_type_name + '.' + output_ext) elif os.path.isdir (options.input_path): bn += '_sumstacks%s' % (nof_stacks) output_path = os.path.join (dn, bn+'_'+output_type_name + '.' + output_ext) else: raise NotImplementedError ('%s is not file nor directory' % (options.input_path)) output_path = fix_path(output_path) print 'Saving new stack to',output_path if output_ext=='tif': ImageStack(new_images, stack.pathinfo, options=options).save(output_path) elif output_ext=='data': from iocbio.microscope.psf import normalize_unit_volume, discretize value_resolution = stack.pathinfo.get_value_resolution() normal_images = normalize_unit_volume(new_images, stack.get_voxel_sizes()) discrete = discretize(new_images / value_resolution) signal_indices = numpy.where(discrete>0) new_value_resolution = value_resolution * normal_images.max() / new_images.max() ImageStack(normal_images, stack.pathinfo, value_resolution = new_value_resolution).save(output_path, zip(*signal_indices)) elif output_ext=='vtk': from pyvtk import VtkData, StructuredPoints, PointData, Scalars vtk = VtkData (StructuredPoints (new_images.shape), PointData(Scalars(new_images.T.ravel()))) vtk.tofile(output_path, 'binary') else: raise NotImplementedError (`output_ext`)
def runner(parser, options, args): smoothness = int(options.smoothness or 1) verbose = options.verbose if options.verbose is not None else True if not hasattr(parser, 'runner'): options.output_path = None if args: if len(args) == 1: if options.input_path: print >> sys.stderr, "WARNING: overwriting input path %r with %r" % ( options.input_path, args[0]) options.input_path = args[0] elif len(args) == 2: if options.input_path: print >> sys.stderr, "WARNING: overwriting input path %r with %r" % ( options.input_path, args[0]) options.input_path = args[0] options.output_path = args[1] else: parser.error( "incorrect number of arguments (expected upto 2 but got %s)" % (len(args))) options.input_path = fix_path(options.input_path) stack = ImageStack.load(options.input_path, options=options) voxel_sizes = stack.get_voxel_sizes() new_images = stack.images.copy() background = (stack.pathinfo.get_background() or [0, 0])[0] print 'Image has background', background window_width = options.window_width or None if window_width is None: dr = stack.get_lateral_resolution() dz = stack.get_axial_resolution() if dr is None or dz is None: window_width = 3.0 scales = tuple( [s / (window_width * min(voxel_sizes)) for s in voxel_sizes]) else: print 'lateral resolution: %.3f um (%.1f x %.1f px^2)' % ( 1e6 * dr, dr / voxel_sizes[1], dr / voxel_sizes[2]) print 'axial resolution: %.3f um (%.1fpx)' % (1e6 * dz, dz / voxel_sizes[0]) vz, vy, vx = voxel_sizes m = 3 scales = (m * vz / dz, m * vy / dr, m * vx / dr) window_width = '%.1fx%.1f' % (dz / m / vz, dr / m / vy) else: window_width = options.window_width scales = tuple( [s / (window_width * min(voxel_sizes)) for s in voxel_sizes]) print 'Window size in pixels:', [1 / s for s in scales] apply_window_inplace(new_images, scales, smoothness, background) if options.output_path is None: b, e = os.path.splitext(options.input_path) suffix = '_window%s_%s' % (window_width, smoothness) options.output_path = b + suffix + (e or '.tif') options.output_path = fix_path(options.output_path) if verbose: print 'Leak: %.3f%%' % (100 * (1 - new_images.sum() / stack.images.sum())) print 'MSE:', ((new_images - stack.images)**2).mean() print 'Energy:', ((stack.images)**2).sum() print 'Saving result to', options.output_path ImageStack(new_images, stack.pathinfo).save(options.output_path)