def ffp_keep(img,frnum,par,prevlist): no_act = prevlist.shape[1] #print "no active in ffp_keep: %i" %no_act keeplist = np.zeros(no_act) keep_num = 0 dimx = img.shape[1] dimy = img.shape[0] #generate background. note img is assumed to already be background subtracted. #instead, want to compare signal / background to decide whether to keep fr_bk = smbkgr.sm_bkgr(img,par.bksize) for i in range(0,no_act): if (fr_bk[prevlist[1,i],prevlist[0,i]] * par.int_thr < img[prevlist[1,i],prevlist[0,i]]): keeplist[i] = 1 #display result img_dispk = np.copy(img) for i in range(0,no_act): if keeplist[i] == 0: for k in range(-5,6): for l in range(-5,6): if circle[k+5,l+5] == 1: img_dispk[prevlist[1,i]+k,prevlist[0,i]+l] = 60 else: for k in range(-5,6): for l in range(-5,6): if circle[k+5,l+5] == 1: img_dispk[prevlist[1,i]+k,prevlist[0,i]+l] = 100 #save img_dispk #im = Image.fromarray(img_dispk.astype(np.uint8)) #im.show() imname = par.file + "_peakfind\\keep_fr"+ str(frnum)#+'.bmp' d=os.path.dirname(imname) if not os.path.exists(d): os.makedirs(d) np2ctimg.saveimg(img_dispk,imname, colortable = ct,autoscale = False) print "based on intensity, keeping %d out of %d spots" %(np.sum(keeplist),no_act) return keeplist
def ffp_slice(img,frnum,par):
dimx = img.shape[1]
dimy = img.shape[0] #not necessarily the same as the camera's whole frame
#bordersize : how much of the edge to ignore
#replaced with par.frameborder to allow easier control
#first, set all values < threshold to zero
med = np.median(np.reshape(img,(dimx*dimy,1)))
thresh = par.std + med
#or, for keep, threshk
threshk = par.std / par.keepratio + med
img_g = np.copy(img) #for finding good peaks
img_k = np.copy(img) #for finding keep peaks
img_disp = np.copy(img) #for display
img_disp = np.rint(img_disp) #to integer values
img_disp = img_disp.astype(int)
#gaussian peak pattern for position refinement
g_peaks = np.zeros((3,3,7,7))
for k in range(0,2):
for l in range(0,2):
offx = -0.5*float(l)
offy = -0.5*float(k)
for i in range(0,7):
for j in range(0,7):
dist = 0.4 * ((float(j) - 3.0 + offx)**2 + (float(i) - 3.0 + offy)**2)
g_peaks[k,l,i,j] = 2.0*math.exp(-dist)
#set to zero all pixels below threshold
for i in range(0,dimx):
for j in range(0,dimy):
if(img_g[j,i] < thresh): img_g[j,i] = 0
if(img_k[j,i]<threshk): img_k[j,i] = 0
good = np.zeros((2,100000))
keep = np.zeros((2,100000))
no_good = 0
no_keep = 0
for i in range(par.frameborder,dimx - par.frameborder+1):
for j in range(par.frameborder,dimy-par.frameborder+1):
if img_k[j,i] > 0:
gp = 0 #flag for peak is good, not just keep
if img_g[j,i] > 0 : gp = 1 #passes std test; check around peak below
#find the nearest maximum
foob = img_k[j-3:j+4,i-3:i+4]
y,x = np.unravel_index(foob.argmax(),foob.shape) #make sure this works as expected
z = foob[y,x]
y-=3
x-=3
#analyze only peaks in the current column
if (y ==0 and x ==0):
y = y+j
x = x + i
#check if it is a good peak - surrounding points near enough zero intensity
quality = 1
for k in range(-5,6):
for l in range(-5,6):
if circle[k+5,l+5] > 0 :
if img[j+k,i+l] > (med+0.5*float(z)): gp= 0 #peak isn't good.
#less picky for keep:
if img[j+k,i+l] > (med+0.5*float(z)*par.keepratio) : quality = 0
if quality == 1 :
#refine peak position to nearest half pixels
#print "pre refine (x,y):"
#print x,y
curbest = 10000
best_x = 1
best_y = 1
for k in range(0,3):
for l in range(0,3):
#how well is the data described by a gaussian at this position. faster than fitting to refine position
diff = np.sum(np.absolute(float(z) *g_peaks[k,l,:,:] - img[y-3:y+4,x-3:x+4]))
if diff < curbest:
best_x = l
best_y = k
curbest = diff
flt_x = float(x) - 0.5*float(best_x-1)
flt_y = float(y) - 0.5*float(best_y-1)
#print 'post refine x,y'
#print flt_x,flt_y
#now, add peak to list if appropriate
if gp == 1:
good[0,no_good] = flt_x
good[1,no_good] = flt_y
no_good = no_good + 1
#at this point, any peak should be added to keep
keep[0,no_keep] = flt_x
keep[1,no_keep] = flt_y
no_keep = no_keep + 1
#'draw' peak where it was found, after refinement
if gp == 1:
for k in range(-5,6):
for l in range(-5,6) :
if circle[k+5,l+5] == 1:
img_disp[flt_y+k,flt_x+l] = 100
elif par.keeptype == 0: #only display these results if they are being used
for l in range(-5,6) :
if circle[k+5,l+5] == 1:
img_disp[flt_y+k,flt_x+l] = 60 #different color for circle
#FIXME: while this does overwrite the existing .bmp, the date/time stamp isn't updated. (at least on my laptop)
#even if the original file is deleted. can fix by deleting the whole folder before rerunning analysis
#im = Image.fromarray(img_disp.astype(np.uint8))# use mode = 'RGB' after converting grayscale to rainbow values in RGB
#im.show()
imname = par.file + "_peakfind\\good_fr"+ str(frnum)#+'.bmp'
d=os.path.dirname(imname)
if not os.path.exists(d):
os.makedirs(d)
np2ctimg.saveimg(img_disp,imname, colortable = ct, autoscale = False)
print "frame: " + str(frnum) + " at " + datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S') + " -- no_good = %d. no_keep = %d" %(no_good,no_keep)
#put good and keep into a list to return together
peaks = [good[:,0:no_good],keep[:,0:no_keep]]
return peaks
