1. Field of the Invention
The disclosure generally relates to noise reduction, and more particularly to a system and method of reducing noise for an image sensor.
2. Description of Related Art
Images taken from a digital camera equipped with image sensor (such as a complementary metal-oxide-semiconductor (CMOS) image sensor) will inevitably pick up noise from a variety of sources. Common noise sources are flicker noise, bad pixel, Gaussian noise, green imbalance, etc. Noise reduction is generally required to remove noise as much as possible.
Conventional methods of reducing noise, however, are commonly devised to overcome one type of noise, and largely overlook distinct characteristics of the noises and ignore features in the image. As a result, conventional methods usually cannot satisfactorily achieve a desirable result.
For the foregoing reasons, a need has thus arisen to propose a novel scheme of reducing noise for an image sensor.
In view of the foregoing, it is an object of the embodiment of the present invention to provide a method of reducing noise for an image sensor such that different noise types may be properly reduced using appropriately suitable schemes, and the noise reduction may be performed in considerations of features in an image.
According to one embodiment, a system of reducing noise includes an image sensor with a color filter array (CFA), a noise reduction device, and a color interpolation device. The color filter array (CFA) placed over the image sensor is configured to output raw data. The noise reduction device is configured to correct the raw data according to distances between an original current pixel and neighboring same-color pixels in a process mask, thereby generating a new current pixel so as to output corrected raw data. The color interpolation device is coupled to receive the corrected raw data to result in full-color data.
Regarding a current pixel of type 0, in step 44A, an interpolation may be performed on neighboring pixels along a smoothest direction to result in a corrected value as a new current pixel. Taking
Regarding a current pixel of type 1, in step 44B, a directional filter may be adopted to perform interpolation on neighboring pixels and the current pixel along a direction of the similar neighboring pixel and the current pixel to result in a corrected value as a new current pixel. Taking
Regarding a current pixel of type n (e.g., n=2 to 8), in step 44C, a grouping-patch-based denoising scheme may be adopted on a portion of similar neighboring pixels and the current pixel to result in a corrected value as a new current pixel. Specifically, according to another aspect of the embodiment, the similar neighboring pixels and the current pixel are first categorized into two groups, i.e., group 0 and group 1.
Afterwards, median or mean operation may be performed on the selected group based on a range of the selected, group, thereby obtaining a corrected value. In the embodiment, the range is defined as an absolute difference between a maximum pixel value and a minimum pixel value of the selected group. If the range is higher than a predetermined value, median operation (median value is used; otherwise, mean operation (mean value) is used.
According to a further aspect of the embodiment, in step 45, the mixing of the new current pixel obtained, from step 44A, 44B or 44C and the original current pixel may be further adjusted according to hue and/or saturation of the pixels to generate full-color data. Hue and saturation information may be obtained from red/green/blue pixels using conventional methods. With respect to hue information, if it is determined that hue (within the process mask) of the original current pixel is substantially different from hue of the new current pixel (for example, an absolute difference therebetween is higher than a predetermined value), the new current pixel is then adjusted (e.g., multiplied) by a smaller strength/weight and the original current pixel is adjusted by a larger strength/weight, and vice versa. In other words, less denoising is performed when large hue variation is detected, and vice versa.
With respect to saturation information, if it is determined that saturation (within the process mask) is low (for example, red, green and blue pixels have average values being substantially close to each other, for example, an absolute difference therebetween is less than a predetermined value), the new current pixel is then adjusted (e.g., multiplied) by a smaller strength/weight and the (original) current pixel is adjusted by a larger strength/weight, and vice versa. In other words, less denoising is performed when low saturation is detected, and vice versa.
Although specific embodiments have been illustrated and described, it will be appreciated by those skilled in the art that various modifications may be made without departing from the scope of the present invention, which is intended to be limited solely by the appended claims.
Number | Name | Date | Kind |
---|---|---|---|
4174522 | Taguchi | Nov 1979 | A |
5896469 | Murakami | Apr 1999 | A |
7477781 | Tanbakuchi | Jan 2009 | B1 |
20020126892 | Gindele et al. | Sep 2002 | A1 |
20050219588 | Yoshida | Oct 2005 | A1 |
20050249417 | Song et al. | Nov 2005 | A1 |
20060233439 | Zhao | Oct 2006 | A1 |
20080007657 | Sakaguchi et al. | Jan 2008 | A1 |
20080137947 | Sawada et al. | Jun 2008 | A1 |
20080273102 | Hyakutake et al. | Nov 2008 | A1 |
20090033773 | Kinoshita et al. | Feb 2009 | A1 |
20090147111 | Litvinov et al. | Jun 2009 | A1 |
20110085729 | Shi et al. | Apr 2011 | A1 |
20110090371 | Cote et al. | Apr 2011 | A1 |
20110091101 | Cote et al. | Apr 2011 | A1 |
20110134292 | Hirai | Jun 2011 | A1 |
20110199542 | Hirai | Aug 2011 | A1 |
20120154646 | Sai et al. | Jun 2012 | A1 |
20120281123 | Hoda et al. | Nov 2012 | A1 |
20130050545 | Komatsu | Feb 2013 | A1 |
20130051665 | Shinozaki | Feb 2013 | A1 |
20130236096 | Cerundolo | Sep 2013 | A1 |
20140002697 | Tai et al. | Jan 2014 | A1 |
20140023268 | Tamura | Jan 2014 | A1 |
Number | Date | Country | |
---|---|---|---|
20140375850 A1 | Dec 2014 | US |