Patent Application
20110058739
1. Field of the Invention
The present invention generally relates to image processing, and more particularly to image contrast enhancement without substantial noise boost.
2. Description of the Prior Art
As the image perception of a viewer is commonly more sensitive to contrast than absolute values themselves, contrast enhancement is one of several very important image processes utilized to improve the quality of an image. However, as the image is subjected to contrast enhancement, noise in the original image tends disadvantageously to be amplified or boosted at the same time, as shown in
In order to relieve this problem, noise reduction is performed prior to the contrast enhancement, as shown in
Referring to
γ(x,y)=f(i(x,y))/i(x,y) (1)
where i(x,y) is the (x,y)th pixel value, and f represents a contrast enhancement function constructed in block 34.
The division operation in the expression (1) is performed by a transform ratio construction block 38 or a divider, the implementation of which requires a large gate count or large number of transistors (to implement the divider). As the gate count is one important factor in deciding the overall cost, it is thus impractical to use Kim's transform ratio construction block 38 in products such as consumer electronic products.
Moreover, Kim needs a reserved memory 32 or buffer to compute a locally smoothed transform ratio by low pass filtering the transform ratios of the input pixels in a local window Wp(x,y). The locally smoothed transform ratio out of block 38 is then multiplied with the input pixel i(x,y) from the reserved memory 32. The required use of this reserved memory 32 adds to the overall cost.
For the reason that such a conventional method or system cannot effectively and, more particularly, economically suppress the noise-boost effect when applying contrast enhancement on images, a need has arisen to propose a novel system and method of image contrast enhancement that is not only capable of suppressing noise boost but also can facilitate economic implementation.
In view of the foregoing, it is an object of the present embodiment to provide a system and method of effectively and economically preventing noise boost in image contrast enhancement.
According to one embodiment, a contrast enhancement unit processes an input pixel with a contrast enhancement function, thereby resulting in (e.g., producing) a contrast-enhanced pixel. A delta unit subtracts the input pixel from the contrast-enhanced pixel, thereby resulting in (e.g., producing) a difference value. A delay unit generates a number of delayed difference values with respect to the difference value. A low-pass filter generates a refined difference value according to the difference value and at least a portion of the delayed difference values. An adding unit adds back the refined difference value to the input pixel, thereby resulting in (e.g., producing) an output pixel.
Referring to
Subsequently, in step 42, the input pixel Y(x,y) is subtracted from the contrast-enhanced pixel f(Y(x,y)) by a delta unit 12, resulting in a difference value or delta δ(x,y):
δ(x,y)=f(Y(x,y))−Y(x,y) (2)
Multiple delayed difference values with respect to the difference value δ(x,y) are generated in step 43 by a delay unit 14.
Afterwards, in step 44, the multiple delayed difference values are fed to a low-pass filter 16 to result in a refined (or smoothed) difference value Rδ(x,y). In the embodiment, the low-pass filter 16 may be implemented by, but is not limited to, a one-dimensional (1D) low-pass filtering expressed as follows:
Rδ(x,y)=[δ(x,y)*6+δ−1(x,y)*5+δ−2(x,y)*3+δ−3(x,y)*2+δ−4(x,y)*2+δ−5(x,y)*b 1+δ−1(x,y)*5+δ2(x,y)*3+δ3(x,y)*2+δ4(x,y)*2+δ5(x,y)*1]/32 (3)
In general, the refined difference value Rδ(x,y) may be expressed as follows:
Rδ(x,y)=[δ(x,y)*w0+δ−1(x,y)*w−1+δ−2(x,y)*w−2+ . . . +δ−n(x,y)*w−n+δ1(x,y)*w1+δ2(x,y)*w2+ . . . +δm(x,y)*wm]/W (4)
where w−n to wm are weightings of the multiple delayed difference values respectively, and W is the sum of the weightings, that is, w=w0+w−1+ . . . +w−n+w1+ . . . +wm, with m and n being integers.
The low-pass filtering is also depicted in
Finally, in step 45, the refined difference value Rδ(x,y) is added back to the input pixel Y(x,y), for example, by an adding unit 18, therefore resulting in an output pixel or the contrast-enhanced pixel without noise boost.
According to the embodiment, the refined difference value Rδ(x,y) is resistant to the noise that is suppressed by the low-pass filter 16. Compared to the conventional scheme such as that disclosed by Kim as discussed above, the divider is no longer needed in the embodiment of the present invention. Accordingly, the associated gate count or the cost may be substantially decreased. Moreover, compared to the scheme disclosed by Kim, the reserved memory or buffer required for computing the locally smoothed transform ratio is no longer needed in the embodiment of the present invention. Accordingly, further costs may be conserved.
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.
