import cv2
import numpy as np
def get_rectangle_corners(image, resize_height, prev_rectangle=None, stability_factor=12):
def process_image(image, height):
ratio = image.shape[0] / height
image = cv2.resize(image, (int(image.shape[1] / ratio), height))
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
gray = cv2.GaussianBlur(gray, (3, 3), 0)
edges = cv2.Canny(gray, 25, 125)
return edges, ratio
def find_rectangle(edges):
contours, _ = cv2.findContours(edges.copy(), cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
contours = sorted(contours, key=cv2.contourArea, reverse=True)[:5]
for contour in contours:
peri = cv2.arcLength(contour, True)
approx = cv2.approxPolyDP(contour, 0.02 * peri, True)
if len(approx) == 4:
return approx
edges, ratio = process_image(image, resize_height)
rectangle = find_rectangle(edges)
if rectangle is not None:
rectangle = rectangle.reshape(4, 2) * ratio
rectangle = [[int(point[0]), int(point[1])] for point in rectangle]
# if prev rectangle exists and the change is smaller than the stability factor, use the previous rectangle
if prev_rectangle is not None:
diff = np.abs(np.array(prev_rectangle) - np.array(rectangle))
if np.all(diff < stability_factor):
return prev_rectangle
return rectangle
else:
return prev_rectangle # if no rectangle detected, use the previous one
prev_rectangle = None
dst = np.float32([[0, 0], [0, 488],[337, 488],[337, 0]])
cap = cv2.VideoCapture(0)
while True:
ret, frame = cap.read()
if ret:
rectangle = get_rectangle_corners(frame, 500, prev_rectangle)
prev_rectangle = rectangle # save the rectangle for the next frame
print(rectangle)
if rectangle is not None:
src = np.float32(rectangle)
m = cv2.getPerspectiveTransform(src, dst)
result = cv2.warpPerspective(frame, m, (337, 488))
cv2.imshow("src", frame)
cv2.imshow("result", result)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()streamlit界面
import streamlit as st
from cnstd import CnStd
from cnocr import CnOcr
import cv2
import numpy as np
import pandas as pd
from PIL import Image
def get_rectangle_corners(image, resize_height, prev_rectangle=None, stability_factor=12):
def process_image(image, height):
ratio = image.shape[0] / height
image = cv2.resize(image, (int(image.shape[1] / ratio), height))
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
gray = cv2.GaussianBlur(gray, (3, 3), 0)
edges = cv2.Canny(gray, 50, 100)
return edges, ratio
def find_rectangle(edges):
contours, _ = cv2.findContours(edges.copy(), cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
contours = sorted(contours, key=cv2.contourArea, reverse=True)[:1]
for contour in contours:
peri = cv2.arcLength(contour, True)
approx = cv2.approxPolyDP(contour, 0.02 * peri, True)
if len(approx) == 4:
return approx
edges, ratio = process_image(image, resize_height)
rectangle = find_rectangle(edges)
if rectangle is not None:
rectangle = rectangle.reshape(4, 2) * ratio
rectangle = [[int(point[0]), int(point[1])] for point in rectangle]
if prev_rectangle is not None:
diff = np.abs(np.array(prev_rectangle) - np.array(rectangle))
if np.all(diff < stability_factor):
return prev_rectangle
return rectangle
else:
return prev_rectangle # if no rectangle detected, use the previous one
# Code where the two scripts are fused together
dst = np.float32([[0, 0], [0, 588],[640, 588],[640, 0]])
prev_rectangle = None
cap = cv2.VideoCapture(0)
std = CnStd()
cn_ocr = CnOcr()
st.sidebar.title("易读写")
option = st.sidebar.selectbox('请选择要运行的功能', ('显示矫正后画面', '图片转文字'))
# 定义两个图像显示窗口
FRAME_WINDOW_RAW = st.image([])
FRAME_WINDOW_TRANSFORMED = st.image([])
# 需要实时显示矫正后画面
if option == '显示矫正后画面':
run = st.checkbox('开始运行')
FRAME_WINDOW = st.image([])
alpha = st.sidebar.slider('亮度调节 Brightness', min_value=0.0, max_value=3.0,
value=1.0) # Add a slider for brightness
angle = st.sidebar.slider('视角调节 Rotation', 0, 360, 0) # Add a slider for rotation
if run:
while True:
ret, frame = cap.read()
if not ret:
st.write("Can't receive frame (Stream end?). Exiting..")
break
rectangle = get_rectangle_corners(frame, 500, prev_rectangle)
prev_rectangle = rectangle # save the rectangle for the next frame
if rectangle is not None:
src = np.float32(rectangle)
m = cv2.getPerspectiveTransform(src, dst)
result = cv2.warpPerspective(frame, m, (640, 588))
raw_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB) # Convert BGR to RGB for raw frame
img_raw = Image.fromarray(raw_frame)
FRAME_WINDOW_RAW.image(img_raw) # Show raw frame
result = cv2.cvtColor(result, cv2.COLOR_BGR2RGB) # Convert BGR to RGB for transformed
result = cv2.convertScaleAbs(result, alpha=alpha, beta=0)
img_transformed = Image.fromarray(result)
FRAME_WINDOW_TRANSFORMED.image(img_transformed) # Show transformed result
else:
st.write('Stopped')
cap.release()
elif option == '图片转文字':
uploaded_file = st.sidebar.file_uploader("选择一个图片文件")
if uploaded_file is not None:
img = Image.open(uploaded_file)
st.image(img, caption='Uploaded Image.', use_column_width=True)
if st.sidebar.button('开始运行'):
# Copy and adapt your code here
with st.spinner('OCR is in progress...'):
np_img = np.array(img) # Convert PIL Image to numpy array
box_infos = std.detect(np_img)
text = []
for box_info in box_infos['detected_texts']:
cropped_img = box_info['cropped_img']
ocr_res = cn_ocr.ocr_for_single_line(cropped_img)
text.append(ocr_res['text'])
st.subheader("Extracted text:")
for line in text:
st.write(line)
st.success('Text extraction complete')