Image_processing_tool

library('keras')
library('tensorflow')
library('devtools') 
library('EBImage')
setwd('C:\Users\vikas\Desktop\footdrop')
pics<-c('f1.jpg','f2.jpg','f3.jpg','f4.jpg','f5.jpg','f6.jpg','nf1.jpg','nf2.jpg','nf3.jpg','nf4.jpg','nf5.jpg','nf6.jpg')
mypics<-list()
for(i in 1:12){
 mypics[[i]]<-readImage(pics[i])
 }
mypics[[1]]
#Resize the pic
for(i in 1:12){mypics[[i]]<-resize(mypics[[i]],28,28)}
str(mypics[[1]])
#Reshape of Images
for(i in 1:12){mypics[[i]]<-array_reshape(mypics[[i]],c(28,28,3)}

for(i in 1:12){mypics[[i]]<-array_reshape(mypics[[i]],c(28,28,3))}

str(mypics)
#Training the model
trainx<-NULL
for(i in 1:5){trainx<-rbind(trainx<-rbind(trainx,mypics[[i]]))}
str(trainx)
 #num [1:5, 1:2352] 0.0457 0.3059 0.8267 0.5494 0.5266 ...
for(i in 7:11){trainx<-rbind(trainx<-rbind(trainx,mypics[[i]]))}
str(trainx)
 num [1:10, 1:2352] 0.0457 0.3059 0.8267 0.5494 0.5266 ...
testx<-rbind(mypics[[6]],mypics[[12]])
trainy<-c(0,0,0,0,0,1,1,1,1,1)
testy<-c(0,1)
#One hot encoding
trainLabels<-to_categorical(trainy)
#creating the model
model<-keras_model_sequential()
testlables<-to_categorical(testy)

#model
model<-kera_model_sequential()

model %>% layer_dense(units=256,activation = 'relu' , input_shape=c(2352)) %>%
+ layer_dense(units=256,activation = 'relu') %>%
+ layer_dense(units=2,activation = 'softmax')

summary(model)

#compile
model %>% complie(loss='binary_crossentropy' , optimizer = optimizer_rmsprop(),metrics=c('accuracy'))
 #fitr model
 history<-model %>% fit(trainx , trainLables , epochs=30 , batch_size=32 , validation_split=0.2)

plot(history)
#Evaluate  and predict the model
model %>% evaluate(trainx , trainlables)

pred<- model %>% predict_classes(trainx)

table(predicted = pred , Actual = trainy)
#probability
 prob<- model %>% predict_proba(trainx)
#confusion matrix
table(predicted = pred , Actual = trainy)