Python DBN dbn.predict use - python-2.7

Hi I am confused with how dbn.predict works I have detected a number in an image using canny and then otsu threshold and resized it to 28x28 etc however when I pass it to dbn.predict I get the following error
ValueError: arrays not aligned for dot product. a dot product was requested of arrays with shapes (1, 28) and (784, 300)
Here is the code for the same, thanks in advance
(cnts, _) = cv2.findContours(edged.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
# sort the contours by their x-axis position, ensuring
# that we read the numbers from left to right
cnts = sorted([(c, cv2.boundingRect(c)[0]) for c in cnts], key = lambda x: x[1])
# loop over the contours
for (c, _) in cnts:
# compute the bounding box for the rectangle
(x, y, w, h) = cv2.boundingRect(c)
# if the width is at least 7 pixels and the height
# is at least 20 pixels, the contour is likely a digit
if w >= 7*scaler and h >= 25*scaler:
# crop the ROI and then threshold the grayscale
# ROI to reveal the digit
roi = gray[y:y + h, x:x + w]
thresh = roi.copy()
T = mahotas.thresholding.otsu(roi)
thresh[thresh > T] = 255
thresh = cv2.bitwise_not(thresh)
#thresh = dataset.deskew(thresh, 28)
#thresh = dataset.center_extent(thresh, (28, 28))
#print thresh.size
#cv2.imshow("Thresh", thresh)
# Resize the image
roi = cv2.resize(thresh, (28, 28), interpolation=cv2.INTER_AREA)
roi = cv2.dilate(roi, (3, 3))
print roi.size
cv2.imshow("ROI", roi)
cv2.waitKey(0)
# Calculate the HOG features
roi_hog_fd = hog(roi, orientations=9, pixels_per_cell=(28, 28), cells_per_block=(1, 1), visualise=False)
nbr = clf.predict(np.array([roi_hog_fd], 'float64'))
pred = dbn.predict(np.atleast_2d(roi/255))

Related

How to match in range color? [duplicate]

Below is my python code for tracking white color objects. It works - but only for a few seconds and then the whole screen turns black and in some times it not work. I experimented with blue color and it works - but white and green are giving me problems:
import cv2
import numpy as np
cap = cv2.VideoCapture(0)
while(1):
_, frame = cap.read()
hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)
# define range of white color in HSV
# change it according to your need !
sensitivity = 15
lower_white = np.array([0,0,255-sensitivity])
upper_white = np.array([255,sensitivity,255])
# Threshold the HSV image to get only white colors
mask = cv2.inRange(hsv, lower_white, upper_white)
# Bitwise-AND mask and original image
res = cv2.bitwise_and(frame,frame, mask= mask)
cv2.imshow('frame',frame)
cv2.imshow('mask',mask)
cv2.imshow('res',res)
k = cv2.waitKey(5) & 0xFF
if k == 27:
break
cv2.destroyAllWindows()
Well, first thing you should know what color space you are using.
Just a small tutorial of color spaces in OpenCV for Mat of type CV_8UC3. (Images from Wikipedia)
HSV
In the HSV (Hue, Saturation, Value) color space, H gives the color dominant color, S the saturation of the color, V the lightness. In OpenCV, the ranges are different. S,V are in [0,255], while H is in [0, 180]. Typically H is in range [0,360] (the full circle), but to fit in a byte (256 different values) it's value is halved.
In HSV space is easier to separate a single color, since you can simply set the proper range for H, and just take care that S is not too small (it will be almost white), and V is not too small (it will be dark).
So for example, if you need almost blue colors, you need H to be around the value 120 (say in [110,130]), and S,V not too small (say in [100,255]).
White is not a hue (the rainbow doesn't have white color in it), but is a combination of color.
In HSV, you need to take all range of H (H in [0, 180]), very small S values (say S in [0, 25]), and very high V values (say V in [230, 255]). This basically corresponds to the upper part of the central axis of the cone.
So to make it track white objects in HSV space, you need:
lower_white = np.array([0, 0, 230])
upper_white = np.array([180, 25, 255])
Or, since you defined a sensitivity value, like:
sensitivity = 15
lower_white = np.array([0, 0, 255-sensitivity])
upper_white = np.array([180, sensitivity, 255])
For other colors:
green = 60;
blue = 120;
yellow = 30;
...
sensitivity = 15
// Change color with your actual color
lower_color = np.array([color - sensitivity, 100, 100])
upper_color = np.array([color + sensitivity, 255, 255])
Red H value is 0, so you need to take two ranges and "OR" them together:
sensitivity = 15
lower_red_0 = np.array([0, 100, 100])
upper_red_0 = np.array([sensitivity, 255, 255])
lower_red_1 = np.array([180 - sensitivity, 100, 100])
upper_red_1 = np.array([180, 255, 255])
mask_0 = cv2.inRange(hsv, lower_red_0 , upper_red_0);
mask_1 = cv2.inRange(hsv, lower_red_1 , upper_red_1 );
mask = cv2.bitwise_or(mask1, mask2)
Now you should be able to track any color!
Instead of having to guess and check the HSV lower/upper color ranges, you can use a HSV color thresholder script to determine the ranges with trackbars. This makes it very easy to define the ranges for whatever color you're trying to segment. Just change the input image in cv2.imread. Example to segment white
import cv2
import numpy as np
def nothing(x):
pass
# Load image
image = cv2.imread('1.jpg')
# Create a window
cv2.namedWindow('image')
# Create trackbars for color change
# Hue is from 0-179 for Opencv
cv2.createTrackbar('HMin', 'image', 0, 179, nothing)
cv2.createTrackbar('SMin', 'image', 0, 255, nothing)
cv2.createTrackbar('VMin', 'image', 0, 255, nothing)
cv2.createTrackbar('HMax', 'image', 0, 179, nothing)
cv2.createTrackbar('SMax', 'image', 0, 255, nothing)
cv2.createTrackbar('VMax', 'image', 0, 255, nothing)
# Set default value for Max HSV trackbars
cv2.setTrackbarPos('HMax', 'image', 179)
cv2.setTrackbarPos('SMax', 'image', 255)
cv2.setTrackbarPos('VMax', 'image', 255)
# Initialize HSV min/max values
hMin = sMin = vMin = hMax = sMax = vMax = 0
phMin = psMin = pvMin = phMax = psMax = pvMax = 0
while(True):
# Get current positions of all trackbars
hMin = cv2.getTrackbarPos('HMin', 'image')
sMin = cv2.getTrackbarPos('SMin', 'image')
vMin = cv2.getTrackbarPos('VMin', 'image')
hMax = cv2.getTrackbarPos('HMax', 'image')
sMax = cv2.getTrackbarPos('SMax', 'image')
vMax = cv2.getTrackbarPos('VMax', 'image')
# Set minimum and maximum HSV values to display
lower = np.array([hMin, sMin, vMin])
upper = np.array([hMax, sMax, vMax])
# Convert to HSV format and color threshold
hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
mask = cv2.inRange(hsv, lower, upper)
result = cv2.bitwise_and(image, image, mask=mask)
# Print if there is a change in HSV value
if((phMin != hMin) | (psMin != sMin) | (pvMin != vMin) | (phMax != hMax) | (psMax != sMax) | (pvMax != vMax) ):
print("(hMin = %d , sMin = %d, vMin = %d), (hMax = %d , sMax = %d, vMax = %d)" % (hMin , sMin , vMin, hMax, sMax , vMax))
phMin = hMin
psMin = sMin
pvMin = vMin
phMax = hMax
psMax = sMax
pvMax = vMax
# Display result image
cv2.imshow('image', result)
if cv2.waitKey(10) & 0xFF == ord('q'):
break
cv2.destroyAllWindows()

How to detect randomly curved non-continuous fixed colored lines from an image?

I am very new in the python openCV. I wanted to find the 7 fixed colored randomly curved lines from an image. The result should be boolean which will give me if the image contains the 7 fixed colored randomly curved lines or not. The sample input image is as below:
I also wanted to find out the non-continuous faint green colored trapezium from the image.
I have written the below code to filter out the specific color from the image but unable to detect the lines & unable to conclude if the image is containing the 7 different lines & trapezium. Below is my sample code for the same:
import cv2
import numpy as np
boundaries = [
(32, 230, 32), # 2 Green lines
(10, 230, 230), # 1 Yellow line
(230, 72, 32), # 1 Blue line
(255, 255, 255), # 2 White lines
(32, 72, 230) # 1 Red line
]
box = [(0, 100, 0), (100, 255, 100)]
image = cv2.imread('testImage5.png')
image = removeBlackBands(image)
# cv2.imshow('Cropped Image', image)
# cv2.waitKey(0)
for row in boundaries:
# create NumPy arrays from the boundaries
row = np.array(row, dtype="uint8")
mask = cv2.inRange(image, row, row)
cv2.GaussianBlur(mask, (5,5), 0)
cv2.imshow('Filtered', mask)
cv2.waitKey(0)
lines = cv2.HoughLinesP(mask, cv2.HOUGH_PROBABILISTIC, np.pi / 180, 50, 50, 100)
if lines is not None:
for x in range(0, len(lines)):
print("line ", x)
for x1, y1, x2, y2 in lines[x]:
print("x1 = {}, y1 = {}, x2 = {}, y2 = {}".format(x1, y1, x2, y2))
cv2.line(image,(x1,y1),(x2,y2),(0,0, 255),2, cv2.LINE_AA)
pts = np.array([[x1, y1], [x2, y2]], np.int32)
cv2.polylines(image, [pts], True, (0, 255, 0))
cv2.imshow('Processed.jpg', image)
cv2.waitKey(0)
# create NumPy arrays from the boundaries
lower = np.array(box[0], dtype="uint8")
upper = np.array(box[1], dtype="uint8")
# find the colors within the specified boundaries and apply
# the mask
mask = cv2.inRange(image, lower, upper)
output = cv2.bitwise_and(image, image, mask=mask)
output = cv2.cvtColor(output, cv2.COLOR_BGR2GRAY)
output = cv2.Canny(output, 100, 150)
# show the images
# cv2.imshow("output", output)
# cv2.waitKey(0)
cv2.destroyAllWindows()
Can somebody help me? Thanks in Advance..!!!
This is the function I wrote for the same.
def detectNodes(self, image, tolerance=0):
"""
Detect the nodes in the image
Algorithm used:
1. Pre-process the image to filter out the required color
2. Convert the pre-processed image to binary image
Args:
image(np.ndarray): Numpy Nd array of image
tolerance: (int): Margin of consideration while extracting color from image. Default: 0
Returns
True upon success, False otherwise
"""
noOfNodesDetected = 0
curveWidth = 2
noOfNodeDetectThreshold = 5
cropH = self.testData["nodalROI"].get("height")
cropW = self.testData["nodalROI"].get("width")
roiImage = ppu.crop(image, cropH, cropW) # Crop node ROI
for color in self.nodalColorBoundaries.keys():
filtered = ImageProc.colorFilter(roiImage, colors=self.nodalColorBoundaries[color], tolerance=tolerance)
bgrImage = ppu.convertColorSpace(filtered, "bgr_to_gray")
thresh = ppu.threshold(bgrImage, 1, "thresh_binary")
logging.info("The shape of image is [{}]".format((thresh.shape)))
height, width = thresh.shape
pointFraction = self.testData.get("pointsToFormEquationOfCurve", None)
points = [int(fraction * height) for fraction in pointFraction]
logging.info("Point using for formulating the equation are [{}]".format(points))
pointFractionEvaluation = self.testData.get("pointsForEvaluationOfCurve", None)
pointsForEvaluation_h = [int(fraction * height) for fraction in pointFractionEvaluation]
logging.info("Point using for Evaluating the equation are [{}]".format(pointsForEvaluation_h))
curve1 = []
curve2 = []
for point in points:
prevX = 0
flag = 0
for w in range(0, width):
if thresh[point][w] == 255:
if (abs(prevX - w)) > curveWidth:
if flag == 0:
curve1.append((point, w))
flag = 1
else:
curve2.append((point, w))
prevX = w
fitter = CurveFitter1D()
if curve2:
logging.error("Second curve detected with color {} having points {}".format(color, curve2))
if curve1:
x1 = [point[0] for point in curve1]
y1 = [point[1] for point in curve1]
logging.qvsdebug("Points using to find the Polynomial with color {} are {} ".format(color, curve1))
fitter._fit_polyfit(x1, y1, 4)
logging.qvsdebug("Coefficient of the Polynomial with color {} are {} ".format(
color, fitter._fitterNamespace.coefs))
else:
logging.error("Points not found with {}".format(color))
return False
pointsForEvaluation_w = [int(round(fitter._predY_polyfit(point))) for point in pointsForEvaluation_h]
logging.qvsdebug(
"Points using for the verification of Polynomial representing curve with color {} are {} ".format(
color, zip(pointsForEvaluation_h, pointsForEvaluation_w)))
counter = 0
for i in range(len(pointsForEvaluation_h)):
if pointsForEvaluation_w[i] + 2 >= width:
continue # Continue if control is reaching to width of iamge - 2
if any(thresh[pointsForEvaluation_h[i]][pointsForEvaluation_w[i] - 2:pointsForEvaluation_w[i] + 3]):
counter += 1
logging.info(
"Out of {} points {} points are detected on the curve for color {}".format(len(pointsForEvaluation_h),
counter, color))
nodeDetectThreshold = int(len(pointsForEvaluation_h) * 0.6)
if counter >= nodeDetectThreshold:
noOfNodesDetected += 1
if noOfNodesDetected >= noOfNodeDetectThreshold:
logging.info("Nodes found in this frame are [%d], minimum expected [%d]" % (
noOfNodesDetected, noOfNodeDetectThreshold))
return True
else:
logging.error("Nodes found in this frame are [%d], minimum expected is [%d]" % (
noOfNodesDetected, noOfNodeDetectThreshold))
return False

openCV - white color filter [duplicate]

Below is my python code for tracking white color objects. It works - but only for a few seconds and then the whole screen turns black and in some times it not work. I experimented with blue color and it works - but white and green are giving me problems:
import cv2
import numpy as np
cap = cv2.VideoCapture(0)
while(1):
_, frame = cap.read()
hsv = cv2.cvtColor(frame, cv2.COLOR_BGR2HSV)
# define range of white color in HSV
# change it according to your need !
sensitivity = 15
lower_white = np.array([0,0,255-sensitivity])
upper_white = np.array([255,sensitivity,255])
# Threshold the HSV image to get only white colors
mask = cv2.inRange(hsv, lower_white, upper_white)
# Bitwise-AND mask and original image
res = cv2.bitwise_and(frame,frame, mask= mask)
cv2.imshow('frame',frame)
cv2.imshow('mask',mask)
cv2.imshow('res',res)
k = cv2.waitKey(5) & 0xFF
if k == 27:
break
cv2.destroyAllWindows()
Well, first thing you should know what color space you are using.
Just a small tutorial of color spaces in OpenCV for Mat of type CV_8UC3. (Images from Wikipedia)
HSV
In the HSV (Hue, Saturation, Value) color space, H gives the color dominant color, S the saturation of the color, V the lightness. In OpenCV, the ranges are different. S,V are in [0,255], while H is in [0, 180]. Typically H is in range [0,360] (the full circle), but to fit in a byte (256 different values) it's value is halved.
In HSV space is easier to separate a single color, since you can simply set the proper range for H, and just take care that S is not too small (it will be almost white), and V is not too small (it will be dark).
So for example, if you need almost blue colors, you need H to be around the value 120 (say in [110,130]), and S,V not too small (say in [100,255]).
White is not a hue (the rainbow doesn't have white color in it), but is a combination of color.
In HSV, you need to take all range of H (H in [0, 180]), very small S values (say S in [0, 25]), and very high V values (say V in [230, 255]). This basically corresponds to the upper part of the central axis of the cone.
So to make it track white objects in HSV space, you need:
lower_white = np.array([0, 0, 230])
upper_white = np.array([180, 25, 255])
Or, since you defined a sensitivity value, like:
sensitivity = 15
lower_white = np.array([0, 0, 255-sensitivity])
upper_white = np.array([180, sensitivity, 255])
For other colors:
green = 60;
blue = 120;
yellow = 30;
...
sensitivity = 15
// Change color with your actual color
lower_color = np.array([color - sensitivity, 100, 100])
upper_color = np.array([color + sensitivity, 255, 255])
Red H value is 0, so you need to take two ranges and "OR" them together:
sensitivity = 15
lower_red_0 = np.array([0, 100, 100])
upper_red_0 = np.array([sensitivity, 255, 255])
lower_red_1 = np.array([180 - sensitivity, 100, 100])
upper_red_1 = np.array([180, 255, 255])
mask_0 = cv2.inRange(hsv, lower_red_0 , upper_red_0);
mask_1 = cv2.inRange(hsv, lower_red_1 , upper_red_1 );
mask = cv2.bitwise_or(mask1, mask2)
Now you should be able to track any color!
Instead of having to guess and check the HSV lower/upper color ranges, you can use a HSV color thresholder script to determine the ranges with trackbars. This makes it very easy to define the ranges for whatever color you're trying to segment. Just change the input image in cv2.imread. Example to segment white
import cv2
import numpy as np
def nothing(x):
pass
# Load image
image = cv2.imread('1.jpg')
# Create a window
cv2.namedWindow('image')
# Create trackbars for color change
# Hue is from 0-179 for Opencv
cv2.createTrackbar('HMin', 'image', 0, 179, nothing)
cv2.createTrackbar('SMin', 'image', 0, 255, nothing)
cv2.createTrackbar('VMin', 'image', 0, 255, nothing)
cv2.createTrackbar('HMax', 'image', 0, 179, nothing)
cv2.createTrackbar('SMax', 'image', 0, 255, nothing)
cv2.createTrackbar('VMax', 'image', 0, 255, nothing)
# Set default value for Max HSV trackbars
cv2.setTrackbarPos('HMax', 'image', 179)
cv2.setTrackbarPos('SMax', 'image', 255)
cv2.setTrackbarPos('VMax', 'image', 255)
# Initialize HSV min/max values
hMin = sMin = vMin = hMax = sMax = vMax = 0
phMin = psMin = pvMin = phMax = psMax = pvMax = 0
while(True):
# Get current positions of all trackbars
hMin = cv2.getTrackbarPos('HMin', 'image')
sMin = cv2.getTrackbarPos('SMin', 'image')
vMin = cv2.getTrackbarPos('VMin', 'image')
hMax = cv2.getTrackbarPos('HMax', 'image')
sMax = cv2.getTrackbarPos('SMax', 'image')
vMax = cv2.getTrackbarPos('VMax', 'image')
# Set minimum and maximum HSV values to display
lower = np.array([hMin, sMin, vMin])
upper = np.array([hMax, sMax, vMax])
# Convert to HSV format and color threshold
hsv = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
mask = cv2.inRange(hsv, lower, upper)
result = cv2.bitwise_and(image, image, mask=mask)
# Print if there is a change in HSV value
if((phMin != hMin) | (psMin != sMin) | (pvMin != vMin) | (phMax != hMax) | (psMax != sMax) | (pvMax != vMax) ):
print("(hMin = %d , sMin = %d, vMin = %d), (hMax = %d , sMax = %d, vMax = %d)" % (hMin , sMin , vMin, hMax, sMax , vMax))
phMin = hMin
psMin = sMin
pvMin = vMin
phMax = hMax
psMax = sMax
pvMax = vMax
# Display result image
cv2.imshow('image', result)
if cv2.waitKey(10) & 0xFF == ord('q'):
break
cv2.destroyAllWindows()

Find mean position of blackpixels using python

I have a binary image and I need to find the mean values of x and y of the black region. These values are calculated for a set of binary images and their mean values of x and y are plotted I don't know how to find this region and calculate their mean values of x and y. Any help is kindly appreciated.
If black pixels are not registered in some data structure, just calculate center of mass for black pixels:
sx = 0
sy = 0
black_cnt = 0
for y in y-range
for x in x-range
if black(x,y)
sx = sx + x
sy = sy + y
black_cnt++
sx = sx / black_cnt
sy = sy / black_cnt
You can obtain the mean positions using the moments of contours.
In order to find the mean you must calculate the first order moments of the contour.
CODE:
#---Read image and obtain threshold---
im = cv2.imread('1.jpg', 1)
img = cv2.cvtColor(im,cv2.COLOR_BGR2GRAY)
ret, thresh = cv2.threshold(img, 120, 255, 1)
#---Obtain contours---
contours, hierarchy = cv2.findContours(thresh, cv2.RETR_TREE,cv2.CHAIN_APPROX_SIMPLE)
cnts = contours
cv2.drawContours(im, contours, -1, (0, 255, 0), 1)
#---Compute the center/mean of the contours---
for c in cnts:
M = cv2.moments(c)
cX = int(M["m10"] / M["m00"])
cY = int(M["m01"] / M["m00"])
print cX
print cY
Values cX and cY have the mean positions of the contours.

Having trouble with cv2.findContours() function

I am unable to find contour and draw a rectangle around it using this video.
Link: http://www.filedropper.com/outcutscaled
Initially, I tried finding contour by simply using the function cv2.findContours() but it was showing some dots. Then i tried applying some morphological transforms but still no luck as contour is not proper.
Can you please tell me what am i doing wrong?
P.S I am using cv2.bitwise_and() to eliminate shadows. One can also use cv2.bitwise_or().
This is my code:
import cv2
def diffImg(t0, t1, t2):
d1 = cv2.absdiff(t2, t1)
d2 = cv2.absdiff(t1, t0)
return cv2.bitwise_and(d1, d2)
cam = cv2.VideoCapture('out_cut_scaled.mp4')
fgbg = cv2.BackgroundSubtractorMOG2(history=50, varThreshold=50)
winName = "Movement Indicator"
cv2.namedWindow(winName, cv2.CV_WINDOW_AUTOSIZE)
# Read three images first:
t_minus = cv2.cvtColor(cam.read()[1], cv2.COLOR_RGB2GRAY)
t = cv2.cvtColor(cam.read()[1], cv2.COLOR_RGB2GRAY)
t_plus = cv2.cvtColor(cam.read()[1], cv2.COLOR_RGB2GRAY)
while True:
result=diffImg(t_minus, t, t_plus)
#frame = cv2.resize(result, (20, 20))
#fgmask = result
fgmask = cv2.blur(result, (10, 10))
fgmask = fgbg.apply(fgmask)
fgmask = cv2.medianBlur(fgmask, 7)
#ret, thresh = cv2.threshold(result, 127, 255, 0)
thresh = cv2.dilate(fgmask, None, iterations=2)
#thresh = cv2.erode(thresh, None, iterations=2)
"""
contours, hierarchy = cv2.findContours(thresh, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)
#cv2.drawContours(thresh, contours, -1, (0, 255, 0), 3)
for contour in contours:
x, y, w, h = cv2.boundingRect(contour)
if w > 10 and h > 10:
cv2.rectangle(thresh, (x, y), (x + w, y + h), (0, 255, 0), 2)
print len(contours)
"""
cv2.imshow(winName, thresh)
cv2.imshow("frame", t)
# Read next image
t_minus = t
t = t_plus
t_plus = cv2.cvtColor(cam.read()[1], cv2.COLOR_RGB2GRAY)
key = cv2.waitKey(10)
if key == 27:
cv2.destroyWindow(winName)
break
OpenCV version: 2.4.13
Python version: 2.7
the function cv2.findContours modifies the input image and draws the contours as dots in the image. May be that is why you are getting only dots.
To get the complete contours, use cv2.drawContours with thickness set to some high value and line type as 8.