I am training CNN model to recognize the face in such way that the model would have only two class. One is "my face", another one is "not my face". It means it will a binary classifier to recognize my face.
I have tried following approaches to make it work.
Took my face as positive and few other faces as negative class. In this approach, I am facing problems with the untrained faces. Some faces which are not trained are recognizing as my face. This problem is called "false positive".
I want to get rid of false positive case. Can anyone suggest me some tricks with the datasets or CNN network that I should use to make it work?
Apart from CNN if any other techniques are available. Please suggest me.
False positives will plague you with this type of classification and your specific objective.
My experience tells me the best solution would be to test a number of different machine learning techniques/models on the same data set until you discover the one that gives the best results.
As Florian H suggested, you could also get a bigger dataset with more faces. Or maybe your face/the feature set is just similar to the other face.
Lastly, I suggest that you go look at a tool called OpenCV and consult their docs on machine learning.
What do mean by "Can't I use some kind of technique that can generate negative image (I don't know such kind of technique, but just want to create this.)". Are you perhaps referring to watershed or thresholding? OpenCV's documentation includes these topics.
Also, giving a more detailed explanation of your chosen features, how you extract, etc, ... makes it far easier for more experienced people to guide you with best practices and techniques so that you can improve your results.