Ask Your Question

Revision history [back]

Problem with opencv_haartraining correct use of image hkjhklj

hi my name is andrea and I'm trying to use opencv_haartraining. I cut 150 images of a particular eyeglass frame. For make this I photographed putting it behind a sheet of white paper and cutting exactly the region of eyeglasses. next I have past the region cut on a white background. Next I have generate 7000 samples with createsamples and I have set it with bg 256 so the write backgroud is take as trasparent. Finished it I have generated the cascade with haartraining, but the xml generated recognizes a lot false-positive. I assumed that setting the white to transparent the image were considered positive only in regions with different colors. my assumption is correct or I need positive images without "holes"? For greater clarification on what the concept will bring a clearer example. If I wanted to recognize a cross using the above technique. the program would recognize only one or because you have to pass a rectangular region it would recognize the cross with the background of bg in the rectangular area given him?

Problem with opencv_haartraining opencv_haartraining: correct use of image hkjhkljimage

hi Hi my name is andrea and I'm trying to use opencv_haartraining. I cut 150 images of a particular eyeglass frame. For make this I photographed putting it behind a sheet of white paper and cutting exactly the region of eyeglasses. next I have past the region cut on a white background. Next I have generate 7000 samples with createsamples and I have set it with bg 256 so the write backgroud is take as trasparent. Finished it I have generated the cascade with haartraining, but the xml generated recognizes a lot false-positive. I assumed that setting the white to transparent the image were considered positive only in regions with different colors. my assumption is correct or I need positive images without "holes"? For greater clarification on what the concept will bring a clearer example. If I wanted to recognize a cross using the above technique. the program would recognize only one or because you have to pass a rectangular region it would recognize the cross with the background of bg in the rectangular area given him?