Distinguishing object and background

A method developed by Tom Asaki to distinguish objects from a background. In these examples, we correct for a low spatial frequency background by removing low frequency components that best minimize the total variation (TV) of the residual. The results show that we can distinguish between object and background because edges contain high frequency components AND are not penalized. The method is not limited to low frequency backgrounds and is robust to noise.

Background — by admin — last modified 2007-05-19 03:52

In the four examples in this folder, the original image is in the top left. The remaining images in the left column were corrected by removing low frequency FFT components with poor results. The images in the right column were corrected by removing low frequency components that best minimize the TV of the residual. In both columns, the subfigures are the results of removing frequencies of order 1, 2 and 3, respectively. We discuss the analysis further here.

Simple gradient — by admin — last modified 2007-05-19 03:52

A circle on a simple low frequency background that spans the color scale.

Third order background — by admin — last modified 2007-05-19 03:52

A circle on a third order background of similar amplitude to the object.

A background of Gaussians — by admin — last modified 2007-05-19 03:52

A circle on a background of gaussians.

A more complex image — by admin — last modified 2007-05-19 03:52

A fish on a simple gradient.