System and method for random noise estimation in a sequence of images
AssigneeQUALCOMM Incorporated
InventorsFrancois Rossignol, Chon Tam Ledinh
Life of this patent
Estimating the term…
Abstract
A system and method for estimating random noise in an image frame or a sequence of image frames are presented. In some embodiments, the method includes performing Global Noise Estimation by comparing current and past filtered frames; converting global noise estimates into local noise estimates using estimated noise parameters based on current input image's local mean intensity; and providing local noise estimates to an adapted generic spatio-temporal filter. A parameter-based noise model is applied in the noise calculation.
Claims14
A noise reduction system for an image frame, comprising:
- a noise estimation module configured to generate noise parameters indicative of a global noise variance from each frame using a parameter-based model by comparing an input frame and a past filtered frame, and calculating a local mean of the input frame, the noise estimation module comprising:
- an absolute difference calculating module that calculates an absolute difference of the input frame and the past filtered frame;
- and
- a histogram generator coupled to the absolute difference calculating module, the histogram generator receiving the calculated local mean and the absolute difference and classifying the absolute difference in a histogram bin according to its local mean;
- a local variance calculator coupled to the noise estimation module, the local variance calculator configured to convert the global noise variance indicated by the generated noise parameters into a local noise variance;
- an adapted spatio-temporal filter that filters the local noise variance along with the input and the past filtered frames and outputs a current filtered frame;
- and
- a frame buffer storing the current filtered frame for use as the next past filtered frame.
A method of performing global random noise estimation in an image frame by a computer, comprising:
- receiving an input image frame;
- receiving a past filtered image frame;
- calculating a local mean of the input image frame;
- calculating an absolute difference of the input and the past filtered frames;
- classifying the absolute difference in a histogram bin according to the local mean of the input image frame;
- generating noise parameters indicative of a global noise variance from each frame using a parameter-based noise model;
- calculating a local noise variance using the noise parameters indicative of a global noise variance
- filtering the local noise variances along with the current and the past filtered frames using an adapted spatio-temporal filter;
- and
- outputting a filtered frame and storing the filtered frame for use as a subsequent past filtered image frame.
A model-based noise reduction system, comprising:
- a noise estimation module configured to generate noise parameters indicative of a global noise variance from a received input frame and a received past filtered frame, the noise estimation module comprising:
- a noise parameter calculator module configured to generate the noise parameters by solving and transforming a parameter-based noise model having the form y =a xb+c, wherein y is a local noise variance, x is a local mean intensity, and a, b, c are the noise parameters;
- a local variance calculator coupled to the noise estimation module, the local variance calculator configured to convert the global noise variance indicated by the generated noise parameters into a local noise variance;
- an adapted spatio-temporal filter that filters the local noise variance along with the input and the past filtered frames and outputs a current filtered frame;
- and
- a frame buffer storing the current filtered frame for use as a next past filtered frame.
Dependencies are inferred from the claim text.