使用python kivy+opencv实现手写数字识别的安卓app。
app下载:
https://pan.baidu.com/s/1l1iqMYmZeHuq-c_mg15F0Q
提取码:lvev
如有问题,欢迎下方评论!
kivy安卓app实现数字识别
代码部分:
from kivy.app import App
from kivy.uix.boxlayout import BoxLayout
from kivy.graphics.texture import Texture
from kivy.uix.camera import Camera
from kivy.uix.image import Image
from kivy.uix.button import Button
from kivy.clock import Clock
from kivy.graphics import Color,Rectangle
import numpy as np
import cv2
from android.permissions import request_permissions, Permission
request_permissions([
Permission.CAMERA,
Permission.WRITE_EXTERNAL_STORAGE,
Permission.READ_EXTERNAL_STORAGE
])
net = cv2.dnn.readNetFromTensorflow('model.pb')
def detect(src,net):
try:
src_copy=src.copy()
gray=cv2.cvtColor(src,cv2.COLOR_RGB2GRAY)
# thred=np.where(gray>140,255,0).astype('uint8')
_,thred=cv2.threshold(gray,100,255,cv2.THRESH_BINARY_INV)
# thred=255-thred
#开闭运算
k = np.ones((5, 5), np.uint8)
#thred=cv.dilate(thred,k)
thred = cv2.morphologyEx(thred, cv2.MORPH_CLOSE, k)
thred=cv2.dilate(thred,k)
# cv2.imshow('a123', thred)
# cv2.waitKey(0)
cnts=cv2.findContours(thred,cv2.RETR_EXTERNAL,cv2.CHAIN_APPROX_SIMPLE)[0]
# print('轮廓数=',len(cnts))
if len(cnts)>100:
return src_copy,thred
nam=[]
for i in cnts:
area = cv2.contourArea(i)
if area<50:
continue
# if aa!=1:
# b=20
# else:b=15
# print('面积=',area)
b=50
if area>=b:
x, y, w, h = cv2.boundingRect(i)
x2=x-10
y2=y-10
w2=w+20
h2=h+20
cv2.rectangle(src_copy,(x2,y2),(x2+w2,y2+h2),(0,255,0),2)
lkuo=thred[y:y+h,x:x+w]
da = max(h, w)
rate = da / 40
ro = cv2.resize(lkuo, (int(w / rate), int(h / rate)))
h, w = ro.shape
t, b = int((43 - h) / 2), 43 - h - int((43 - h) / 2)
l, r = int((43 - w) / 2), 43 - w - int((43 - w) / 2)
ro = cv2.copyMakeBorder(ro, t, b, l, r, cv2.BORDER_CONSTANT, value=0)
ro = cv2.resize(ro, (40, 40))
ro = np.where(ro > 0, 255, 0).astype('float32')
# cv.imshow('a123', ro)
# cv.waitKey(0)
ro = ro / 255
# ro=np.reshape(ro,(1,40,40,1))
blob=ro.reshape((1,1,40,40))
net.setInput(blob)
out=net.forward()
#图上标出数字
number=np.argmax(out,axis=-1)
nam.append((number[0],x2, y2-5,h/40))
#cv.putText(src_copy, str(number[0]), (x2, y2-5), cv.FONT_HERSHEY_SIMPLEX, h/40, (0, 0, 255), 2)
#print('opencv预测结果:',np.argmax(out,axis=-1),'tensorflow预测结果:',np.argmax(out2,axis=-1))
if len(nam)==0:
return src_copy,thred
for ii in nam:
cv2.putText(src_copy, str(ii[0]), (ii[1], ii[2]), cv2.FONT_HERSHEY_SIMPLEX, ii[3], (0, 0, 255), 2)
return src_copy,thred
except:
return src,src
class MyLayout(BoxLayout):
def __init__(self):
super(MyLayout, self).__init__()
self.orientation='vertical'
self.image1 = Image()
self.image1.color=1,1,1,0
self.image1.allow_stretch=True
self.btn=Button(text='OPEN CAMERA')
with self.canvas:
Color(1,1,1,1)
self.rect=Rectangle(pos=self.pos,size=self.size,source='11.jpg')
self.bind(pos=self.update,size=self.update)
self.padding=[0,0,0,10]
self.spacing=10
self.btn.size_hint=.3,.06
self.btn.pos_hint={'center_x':0.5,'center_y':.5}
self.btn.background_color=1,1,1,.8
# self.btn.size_hint_y=.1
# self.btn.size_hint_x=.2
self.btn.bind(on_press=self.press_btn)
self.camera = Camera(resolution=(640, 480))
self.camera.play=False
self.add_widget(self.image1)
self.add_widget(self.btn)
Clock.schedule_interval(self.pr, 0)
def update(self,*args):
self.rect.pos=self.pos
self.rect.size=self.size
def press_btn(self,arg):
self.camera.play=not self.camera.play
if self.camera.play:
self.image1.color=1,1,1,1
self.btn.text='CLOSE CAMERA'
else:
self.image1.color = 1, 1, 1, 0
self.btn.text = 'OPEN CAMERA'
def pr(self, nap):
if self.camera.texture==None:
return
img_rgb=self.texture_to_numpy(self.camera.texture)[:,:,:3]
img_rgb_90=np.rot90(img_rgb,-1)
img_rgb_zy=img_rgb_90[:,::-1,:]
out,_=detect(img_rgb_zy,net)
#cv2.putText(img_rgba_copy, str(self.counter), (100, 100), cv2.FONT_HERSHEY_SIMPLEX, 2, (0, 255, 0), 3)#
texture1=self.numpy_to_texture(out)
self.image1.texture=texture1
# 将Texture格式的图片对象转换为numpy
def texture_to_numpy(self, data):
image = np.asarray(bytearray(data.pixels), dtype='uint8').reshape((data.height, data.width, 4))
# r_chanel = np.copy(image[:, :, 0])
# g_chanel = np.copy(image[:, :, 1])
# b_chanel = np.copy(image[:, :, 2])
# image[:, :, 0] = b_chanel
# image[:, :, 1] = g_chanel
# image[:, :, 2] = r_chanel
#返回rgba图像
return image
# 将numpy格式图片对象转为Texture
def numpy_to_texture(self, frame):
# frame=cv2.imread("ddd.jpg")
buf1 = cv2.flip(frame, 0)
buf = buf1.tostring()
image_texture = Texture.create(size=(frame.shape[1], frame.shape[0]), colorfmt='rgb')
image_texture.blit_buffer(buf, colorfmt='rgb', bufferfmt='ubyte')
return image_texture
class MyApp(App):
def build(self):
return MyLayout()
if __name__ == '__main__':
MyApp().run()
