Datasets
Models
Downloads: 1
109 lines (109 with data), 13.7 kB
úž_ä> Ń @ s└ d d l m Z d d l j Z d d l Z d d l Z d d l Z d Z d Z
d Z y& d d l Z d d l m
Z
e é Wn e k
rĽ d Z Yn Xe s═ d Z
y d d l Z Wn e k
r╠ d Z
Yn Xe r(e
r(d Z y d d l Z d d l m Z Wn$ e k
r'd Z e d â é Yn XGd d ä d e â Z Gd
d ä d e â Z Gd d
ä d
e â Z Gd d ä d e â Z d d l Z d d ä Z d d d ä Z d d d d d ä Z d S)Ú )┌divisionNTF)┌Image)┌miscz couldn't load ANY image libraryc @ sI e Z d Z d Z d d d ä Z d d ä Z d d ä Z d d
ä Z d S) ┌ _OtsuPyramidz«segments histogram into pyramid of histograms, each histogram
half the size of the previous. Also generate omega and mu values
for each histogram in the pyramid.
Ú c C sń t | â t j k s5 t d t t | â â d â é | j d k rP t d â é | | _ t j | | â \ } } d d ä | Dâ } | j | â \ } } } } d d ä t | â Dâ | _
d d ä t | â Dâ | _ | | _ d S)
z$ bins is number of intensity levels z must be passed numpy array. Got z insteadÚ z4image must be greyscale (and single value per pixel)c S s g | ] } t | â Ĺ q Sę )┌int)┌.0┌hr r ˙I/raid/data/wangqiuli/Documents/YH_Interpretable/Version2/features/otsu.py˙
<listcomp>4 s z+_OtsuPyramid.load_image.<locals>.<listcomp>c S s g | ] } | Ĺ q Sr r )r
┌omegasr r r r
8 s c S s g | ] } | Ĺ q Sr r )r
┌musr r r r
: s N)
┌type┌np┌ndarray┌
ValueError┌str┌ndim┌im┌ histogram┌$_create_histogram_and_stats_pyramids┌reversed┌ omegaPyramid┌ muPyramid┌ ratioPyramid) ┌selfr ┌bins┌hist┌rangesZhistPyrZomegaPyrZmuPyrZratioPyrr r r ┌
load_image% s
z_OtsuPyramid.load_imagec
s t ł â } d ë t t j | ł â â } g } g } g } g } xf t | â D]X } | j ł â ç ç f d d ć t d | ł â Dâ } | ë | ł } | j ł â qO Wd | d <x@ | D]8 ë | j ł â \ }
} } | j |
â | j | â q╝ W| | | | f S)a│ Expects hist to be a single list of numbers (no numpy array)
takes an input histogram (with 256 bins) and iteratively
compresses it by a factor of 2 until the last compressed
histogram is of size 2. It stores all these generated histograms
in a list-like pyramid structure. Finally, create corresponding
omega and mu lists for each histogram and return the 3
generated pyramids.
Ú c s* g | ] } t ł | | ł ů â Ĺ q Sr )┌sum)r
┌i)r ┌ratior r r
Q s zE_OtsuPyramid._create_histogram_and_stats_pyramids.<locals>.<listcomp>r Ú )┌lenr ┌math┌log┌range┌append┌(_calculate_omegas_and_mus_from_histogram)
r r r ┌
reductionsZcompressionFactorZ histPyramidr r ┌_Z reducedHistr r ┌muTr )r r% r r = s&
(
z1_OtsuPyramid._create_histogram_and_stats_pyramidsc C s┬ | j | â \ } } t | â } t d â } g } x/ t | â D]! } | | | 7} | j | â q@ Wt d â } g } x/ t | â D]! } | | | 7} | j | â qä Wt | â }
| | |
f S)zk Comput histogram statistical data: omega and mu for each
intensity level in the histogram
r )┌ _calculate_histogram_pixel_statsr' ┌floatr* r+ ) r r ZprobabilityLevelsZ
meanLevelsr Zptotalr r$ Zmtotalr r/ r r r r, b s z5_OtsuPyramid._calculate_omegas_and_mus_from_histogramc si t ł â } t t ł â â ë ç ç f d d ć t | â Dâ ë ç f d d ć t | â Dâ } ł | f S)a Given a histogram, compute pixel probability and mean
levels for each bin in the histogram. Pixel probability
represents the likely-hood that a pixel's intensty resides in
a specific bin. Pixel mean is the intensity-weighted pixel
probability.
c s g | ] } ł | ł Ĺ q Sr r )r
r$ )┌Nr r r r
ł s zA_OtsuPyramid._calculate_histogram_pixel_stats.<locals>.<listcomp>c s g | ] } | ł | Ĺ q Sr r )r
r$ )┌hist_probabilityr r r
Ő s )r' r1 r# r* )r r r Z
pixel_meanr )r2 r r3 r r0 z s
" z-_OtsuPyramid._calculate_histogram_pixel_statsN)┌__name__┌
__module__┌ __qualname__┌__doc__r! r r, r0 r r r r r s
% r c @ sI e Z d Z d Z d d ä Z d d ä Z d d ä Z d d d
ä Z d S) ┌OtsuFastMultithresholdz╦Sacrifices precision for speed. OtsuFastMultithreshold can dial
in to the threshold but still has the possibility that its
thresholds will not be the same as a naive-Otsu's method would give
c sű g | _ | j | â } t | j | â | _ | j | â } | j d } xŚ t | t | j â â D]} } | j | } | j | } t | | | â } | j | â } | j j
| â | j | ë ł } ç f d d ć | Dâ } qc Wç f d d ć | Dâ S)Nr" c s g | ] } | ł Ĺ q Sr r )r
┌t)┌scalingr r r
Č s zAOtsuFastMultithreshold.calculate_k_thresholds.<locals>.<listcomp>c s g | ] } | ł Ĺ q Sr r )r
r9 )r: r r r
░ s ) Z
threshPyramid┌ _get_smallest_fitting_pyramidr' r r ┌_get_first_guess_thresholdsr* r ┌_ThresholdHunter┌%find_best_thresholds_around_estimatesr+ r ) r ┌k┌start┌
thresholds┌deviater$ r r Zhunterr )r: r ┌calculate_k_thresholdsĽ s
z-OtsuFastMultithreshold.calculate_k_thresholdsc C s: x3 t | j â D]" \ } } t | â | k r | Sq Wd S)zXReturn the index for the smallest pyramid set that can fit
K thresholds
N)┌ enumerater r' )r r? r$ Zpyramidr r r r; ▓ s z4OtsuFastMultithreshold._get_smallest_fitting_pyramidc sů | d } | j d ë ç f d d ć t | d d â Dâ ł g ç f d d ć t d | â Dâ } | j | j d â | d | ů S) z╗Construct first-guess thresholds based on number of
thresholds (k) and constraining intensity values. FirstGuesses
will be centered around middle intensity value.
r" c s g | ] } ł | Ĺ q Sr r )r
r$ )┌midwayr r r
┴ s zFOtsuFastMultithreshold._get_first_guess_thresholds.<locals>.<listcomp>r r& c s g | ] } ł | Ĺ q Sr r )r
r$ )rE r r r
┬ s NÚ )r r* r+ )r r? ZkHalfZ firstGuessesr )rE r r< ║ s
)#z2OtsuFastMultithreshold._get_first_guess_thresholdsNc s4 | d k r | j } t | â ë d g | d g } d g ç f d d ć t d ł d â Dâ d g } t j | d t j â} t j | j d t j â} xĹ t ł d â D] } | | } t j | j d t j â} | rŕ | | k } | | d } | r| | | k M} | | }
| |
} | | 7} qş W| S)Nr c s( g | ] } t d ł | d â Ĺ q S)r r& )r )r
r$ )r? r r r
╬ s zDOtsuFastMultithreshold.apply_thresholds_to_image.<locals>.<listcomp>r& Ú ┌dtype)
r r' r* r ┌array┌uint8┌zeros┌shape┌ones┌bool8) r rA r ZbookendedThresholdsZ
greyValuesZ
finalImager$ ZkSmall┌bwZkLargeZ greyLevelZ greyImager )r? r ┌apply_thresholds_to_image╚ s& 4
z0OtsuFastMultithreshold.apply_thresholds_to_image)r4 r5 r6 r7 rC r; r< rP r r r r r8 Ć s
r8 c @ sI e Z d Z d Z d d d ä Z d d ä Z d d ä Z d d
ä Z d S) r= z^Hunt/deviate around given thresholds in a small region to look
for a better threshold
r" c C s. t | | â | _ | j j | _ | | _ d S)N)┌_BetweenClassVariance┌sigmaBr rB )r r r rB r r r ┌__init__Ú s z_ThresholdHunter.__init__c C sŽ d | d d ä | Dâ f } | } d } xw | j | d | j â D]] } | j j | â } | | k rć t | â t | â k r× | } qA | | k rA | } | } qA W| S)zvGiven guesses for best threshold, explore to either side of
the threshold and return the best result.
r c S s g | ] } d Ĺ q S)r r )r
r9 r r r r
§ s zJ_ThresholdHunter.find_best_thresholds_around_estimates.<locals>.<listcomp>)┌ _jitter_thresholds_generatorr rR ┌get_total_variancer# )r ┌estimatedThresholdsZ bestResults┌bestThresholdsZ bestVariancerA ┌variancer r r r> s
z6_ThresholdHunter.find_best_thresholds_around_estimatesc sp d d ä | Dâ } t | â d k r2 ł j | â Sç f d d ć } t j j | | â } d d ä | Dâ } | S)z~Experimental threshold hunting uses scipy optimize method.
Finds ok thresholds but doesn't work quite as well
c S s g | ] } t | â Ĺ q Sr )r )r
r? r r r r
s zW_ThresholdHunter.find_best_thresholds_around_estimates_experimental.<locals>.<listcomp>Ú
c s! d ł j j d d ä | Dâ â S)Nr& c S s g | ] } t | â Ĺ q Sr )r )r
r? r r r r
s zi_ThresholdHunter.find_best_thresholds_around_estimates_experimental.<locals>.<lambda>.<locals>.<listcomp>rF )rR rU )┌x)r r r ┌<lambda> s zU_ThresholdHunter.find_best_thresholds_around_estimates_experimental.<locals>.<lambda>c S s g | ] } t | â Ĺ q Sr )r )r
r? r r r r
s )r# r> ┌scipy┌optimize┌fmin)r rV Zfxn_to_minimizerW r )r r ┌2find_best_thresholds_around_estimates_experimental s zC_ThresholdHunter.find_best_thresholds_around_estimates_experimentalc
c s | d } t | â d k rq x˛ t | j | j d â D]3 } | | } | | k s7 | | k rb q7 | g Vq7 Wná | d d ů } t | â } xü t | j | j d â D]e } | | } | | k sĘ | | | k rÎ qĘ | j | | d | â } x | D] } | g | Vq¸ WqĘ Wd S)Nr r& )r' r* rB rT )
r rA ┌min_┌max_Z
pastThresh┌offset┌thresh┌mZrecursiveGeneratorZotherThresholdsr r r rT s"
!
!
z-_ThresholdHunter._jitter_thresholds_generatorN)r4 r5 r6 r7 rS r> r_ rT r r r r r= ń s
r= c @ s4 e Z d Z d d ä Z d d ä Z d d ä Z d S)rQ c C s4 | | _ | | _ t | â | _ t | â | _ d S)N)r r r' r r# ┌muTotal)r r r r r r rS ; s z _BetweenClassVariance.__init__c C sz d g | | j d g } t | â d } d } xB t | â D]4 } | | } | | d } | | j | | â 7} q> W| S)zéFunction will pad the thresholds argument with minimum and
maximum thresholds to calculate between class variance
r r& )r r' r* ┌ _between_thresholds_variance)r rA Z
numClasses┌sigmar$ ┌k1┌k2r r r rU B s
z(_BetweenClassVariance.get_total_variancec C sI | j | | j | } | j | | j | } | j } | | | d S)z6to be usedin calculating between-class variances only!r" )r r re )r rh ri ┌omega┌mur/ r r r rf Q s z2_BetweenClassVariance._between_thresholds_varianceN)r4 r5 r6 rS rU rf r r r r rQ 9 s rQ c C s d } d } x} t | j d â D]h } x_ t | j d â D]J } | | | | k rb | | | } | | | | k r: | | | } q: Wq Wt | | â x^ t | j d â D]I } x@ t | j d â D]+ } t j | | | â ră d | | | <qă Wqş W| | | | } | d S)Nr& i' r g Óo@rF )r* rL ┌printr( ┌isnan)┌img┌Max┌Minr$ ┌jr r r ┌ normalizeZ s
rr r& c C s t | j â â S)zŮ
ret_num: the number of image return
img_nums: the number of calculate thresholds
categories_pixel_nums: the number of pixel categories
return the images with only categories_pixel_nums types of pixels
)rr ┌copy)rn Zcategories_pixel_numsr r r ┌_otsul s rt g Ó?c C s, t â } | j | â | j | | g | â S)N)r8 r! rP )rn ┌upper┌down┌
categoriesZotr r r ┌ otsu_helperä s
rx g Ó┐)┌
__future__r ┌matplotlib.pyplot┌pyplot┌pltr( ┌numpyr ┌cv2ZUSE_PILZUSE_CV2Z USE_SCIPY┌PILr ┌ ImportErrorr\ r ┌ RuntimeError┌objectr r8 r= rQ rr rt rx r r r r ┌<module> sF
pUU