How to improve the accuracy of points detected ?

[henriquem]

Somebody knows how I can improve to become points found more accuracy ?

For Example in example here,
the pre-processing of image is just convert image to grayscale

There is something more that you do for sharing with us ?

I did cvEqualizeHist(gray, gray); and reduce specks from image:

cvMorphologyEx(gray, gray, null, null, CV_CAP_OPENNI, 1);

I tried to applied Gabor Filter but the accuracy of points wasn't good.

[saudet]

Which function are you using to detect points?

[henriquem]

the same used by example:
// only gray scale
final IplImage input= cvCreateImage(cvGetSize(image), IPL_DEPTH_8U, 1);
cvCvtColor(image, imageBW, CV_BGR2GRAY);

BytePointer elem = cvGetSeqElem(rects, iface);
CvRect rect = new CvRect(elem);

            bbox[0] = rect.x();
            bbox[1] = rect.y();
            bbox[2] = rect.x() + rect.width();
            bbox[3] = rect.y() + rect.height();

            flandmark_detect(input, bbox, model, landmarks);

[saudet]

It looks like this is detecting corners, so we might be able to get subpixel accuracy by calling cvFindCornerSubPix() on the detected corners.