TWI438718B - Image processing method and system by using adaptive inverse hyperbolic curve - Google Patents

Description

Adaptive anti-hyperbolic image processing method and system thereof

The present invention relates to an image processing method, and in particular to a method for processing image brightness.

The requirement of a general digital camera is to maintain the brightness of the main target in the focal length, such as the brightness distribution of the face area. According to this requirement, most digital cameras use gamma functions as the basis for image enhancement. However, the use of such image enhancement methods often results in a deterioration of the contrast between the main target and the background.

In the field of image contrast enhancement technology, the gamma function is a commonly used method. However, this method has two disadvantages: First, the image with too much brightness distribution during image contrast enhancement (eg: Backlit images, over-bright images, and over-dark images cannot preserve the details of the brightness distribution of the original image and are prone to distortion. Second, this algorithm can only achieve global contrast enhancement of the image and can not achieve regional contrast enhancement, and can not meet the human visual response curve of brightness, and make the image unsmooth or distorted. The phenomenon.

Therefore, one aspect of the present disclosure is to provide an adaptive inverse hyperbolic image processing method to meet human visual response curves for brightness.

According to an embodiment of the present invention, an adaptive inverse hyperbolic image processing method is provided, including the following steps: First, step a is performed to extract an original luminance data from an input image. Then, step b is performed to set an adaptive inverse hyperbolic function according to the average value and the standard deviation of the original luminance data. Next, step c is performed to correct the original luminance data by using an adaptive inverse hyperbolic function to generate an adaptive luminance data. Finally, step d is performed to correct the input image with adaptive brightness data to generate an output image.

In addition, according to other embodiments of the present technical aspect, step a may be to convert the three primary color red green blue (RGB) data formats of the input image through color space conversion to obtain brightness data; and step d may replace the original brightness data with adaptive brightness data. It is worth noting that if the original luminance data of the input image has a wide or extreme range of distribution, step b can divide the original luminance data into multiple band luminance data, and then calculate the average and standard of the luminance data of the respective frequency bands. Poor, correspondingly set a plurality of adaptive anti-hyperbolic functions to correspond to the luminance data of these bands one by one. On the other hand, step b can calculate the adaptive deviation parameter ( bias ) of an area according to the original brightness data or the average value of the brightness data of each frequency band, and calculate the adaptability of a region according to the original brightness data or the standard deviation of the brightness data of each frequency band. The gain parameter ( gain ) is then used to set the adaptive inverse hyperbolic function using the adaptive bias parameter ( bias ) and the adaptive gain parameter ( gain ).

Another aspect of the present disclosure is to provide an adaptive inverse hyperbolic image processing system for performing the adaptive inverse hyperbolic image processing method described above.

According to an embodiment of the present invention, an adaptive inverse hyperbolic image processing system is provided, including an input unit, an image processing unit, and an output unit. The input unit is configured to obtain an image; the image processing unit is configured to receive the image, and then perform the adaptive inverse hyperbolic image processing method; and the output unit is configured to output the processed output image.

Thereby, the adaptive anti-hyperbolic image processing method and system thereof of the above embodiments can satisfy the human visual response to the brightness contrast in the image brightness contrast processing, and avoid the phenomenon that the image is not smooth or distorted.

According to medical related data, there are many receiving cells for light and color distributed in the retinal structure of our human eyes. The receiving cells can be divided into two types: (1) Rod cells, (2) cones. Conne cell. Rod cells are about 130 million in the retina of human eyes. They are quite sensitive to low light, but this cell does not have the ability to distinguish colors, mainly in low light. Conversely, under strong light, the rod cells will be less sensitive to light due to the decrease of the photopigment, which is called light adaptation. Cone cells have about six to seven million retinas distributed in the human eye, and their cells are stimulated by intense light and shades, which act primarily when watching anything during the day.

In the real world, the intensity distribution of brightness has a large range. According to the response curve of the human visual system, the low-scale end is that the starlight belongs to extremely weak light, and the intensity of the average brightness is about 10 ^-3 cd/cm ² , and the intensity of the average brightness during the day is about 10 ⁵ cd/ Cm ² belongs to high glare. However, the visible range of the human eye is only between 1 and 10 ⁴ cd/cm ² ; therefore we almost lose all the details in the dark and the strong light in the visible spectrum. Due to the human visual sense, the human brain has the function of differential complement calculation and image enhancement. It is comparable to any digital camera, camera or display on the market today, and it also has a fully automatic non-segment connection. With the near-perfect post-production capability, it is better to see the image seen by the human eye and the brightness information it receives than to obtain the image through the image capture device, and then display the image by the display.

The main purpose of image contrast enhancement technology is to enhance the visual effect of the image, making it easy for the human eye to recognize the image or make the image clearer and more detailed, and to make the computer machine easy to analyze the image data and identification, and has the visual sense function like human being. However, according to our observations, most of the image capture devices currently sold on the market, such as digital cameras and cameras, have adopted gamma functions to make image contrast enhancement adjustments to the digital images captured by them. This method of processing will result in a contrast between the main target and its background contrast information after contrast enhancement, and usually cannot meet the human visual response curve of brightness contrast; and it is inconsistent with the state where the brightness distribution of the image is too extreme. Human visual needs. Therefore, the present disclosure proposes an image processing method that can be easily used and can improve digital image contrast to overcome the problems we have observed above.

For the analysis of the principle, please refer to the first picture and the second picture together. The first picture shows the spectrum of the brightness distribution of various input images (the result of histogram statistics), and the second picture shows the various input images corresponding to the first picture. The human eye best fits the contrast adjustment curve. In general, the contrast distribution of images that we can capture in our daily life using various devices can be divided into five types as shown in Figure 1: dark image, bright image (bright image) , back-lighted image, low contrast image and high contrast image. In the dark image, the brightness distribution is only distributed in the dark area. The brightness distribution of the over-bright image is only distributed in the bright area, and the brightness of the backlight image is distributed in the dark area and the bright area. The low-contrast image brightness distribution is concentrated in the In the middle area, high-contrast image brightness is evenly distributed throughout the spectrum.

As mentioned above, in order to meet the visual response curve of human eye contrast, different conversion curves should be used for images with different brightness distributions. As shown in Figure 2, we use medical information to find out the different contrast curves that should be used for different types of contrast images, and to construct their mathematical models. In general, an image that is too dark means that the average value of the original luminance data ( mean ) is less than 0.5, and the corresponding curve as shown in Fig. 2(a) should be used. For an image that is too bright, meaning that the average of the original brightness data is greater than 0.5, the corresponding curve as shown in Figure 2 (b) should be used. For the original brightness data of the image distributed at both ends of the dark area and the bright area, the corresponding curve as shown in Fig. 2(c) should be used. The original brightness data for the image is concentrated in the middle region, and the average value is approximately equal to 0.5. The corresponding curve as shown in Fig. 2(d) should be used. For the image raw luminance data to be evenly distributed throughout the spectrum, the corresponding curve as shown in Fig. 2(e) is used.

Therefore, the adaptive anti-hyperbolic image processing method of the present embodiment establishes an adjustment mechanism to define different reaction curves for different brightness distributions, and then apply appropriate contrast adjustment according to the input image. Thereby, the embodiment can be constructed on the human visually intuitive perception, satisfying the human visual response curve of brightness, and avoiding the phenomenon that the image is not smooth or distorted.

Please refer to FIG. 3, which is a flow chart of the steps of the adaptive inverse hyperbolic image processing method of the present embodiment. In FIG. 3, the adaptive inverse hyperbolic image processing method of the present embodiment includes the following steps: First, as shown in step 101, an original luminance data is taken out from an input image. Then, as shown in step 102, an adaptive inverse hyperbolic function is set based on the average and standard deviation of the original luminance data. Next, as shown in step 103, the original luminance data is corrected using an adaptive inverse hyperbolic function to generate an adaptive luminance data. Finally, as shown in step 104, the input image is corrected using adaptive brightness data to produce an output image. In the present embodiment, in step 102, the adaptive deviation and the adaptive gain are calculated mainly by the average value and the standard deviation of the original luminance data, and then the adaptive inverse hyperbolic function is set.

Technically, please refer to FIG. 4, which is a flow chart of steps of another embodiment of the present disclosure. First, as shown in step 210, the value of the input image is first converted from the original format to the RGB value of the floating point representation, and the converted color image is converted from the original image format into a color space (for example, RGB to HSV or RGB to HSI), and as shown in steps 220 and 230, raw luminance data and raw chroma data (such as chrominance and saturation) are obtained. Then, as shown in step 240, adaptive adjustment is performed according to the original luminance data, for example, the adaptive deviation and the adaptive gain are calculated according to the average value and the standard deviation of the original luminance data, and the anti-hyperbolic tangent function curve after the adaptive adjustment is defined. Then, the contrast is adjusted by the adaptive inverse hyperbolic tangent function curve. Next, as shown in step 250, the adjusted adaptive brightness data is integrated with the previously separated original chroma data, for example, the original brightness data is directly replaced by the adaptive brightness data. Finally, as shown in step 260, the image in the image color format is converted into an image in red, green, and blue (RGB) format, thereby generating an adjusted output image.

Please refer to FIG. 5 again. FIG. 5 is a detailed step flow chart of step 240 of FIG. 4. Specifically, as shown in steps 241 and 242, the present embodiment first calculates the average and standard deviation of the original luminance data, that is, the type of the original luminance data that is spectrally distributed. Then, as shown in steps 243 and 244, an adaptive deviation parameter ( bias ) is calculated using the average value, and an adaptive gain parameter ( gain ) is calculated using the standard deviation. Finally, as shown in step 245, the adaptive inverse hyperbolic function is set using the adaptive bias parameter ( bias ) and the adaptive gain parameter ( gain ), that is, the contrast conversion curve most suitable for this type of luminance distribution is generated.

As described above, the present embodiment takes an inverse hyperbolic tangent function as an example to teach how to set the adaptive deviation function bias ( x ) and the adaptive gain function gain ( x ) to define an adaptive inverse hyperbolic function. First, please refer to Figure 6, which is the processing curve of various types of images that the adaptive anti-hyperbolic tangent function can correspond to. The form of this function conforms to the data obtained from the measurements, as well as the reflected response from the electrical perception of various different kinds of light. It also provides a good current physiology and psychometric measurement of human visual function. To enhance the contrast of the image, we use the inverse hyperbolic tangent function as shown in the following equation (1). The algorithm adds the parameters of the adaptive deviation function bias ( x ) and the adaptive gain function gain ( x ) to control the curve shape of the inverse hyperbolic tangent function, as shown in the following equation (2):

From this we can see that the curve in the middle has a characteristic similar to linear.

Please refer to Figure 7, which is a displacement curve of the adaptive inverse hyperbolic tangent function. The adaptive anti-hyperbolic tangent function correspondence curve is an algorithm based on the human eye's response curve to Luminance, and can be adapted to the characteristic function of the brightness distribution of various contrast type images. Maps for very small or very large brightness values, which enhances contrast between dark and bright areas. This function has a one-time asymptotic feature, which means that the range of the output map is always between 0 and 1. Another advantage of this function is that the tangent function provides an inverse hyperbola to map too dark and too bright details.

Among them, Fig. 7(a) is a graph of the inverse hyperbolic tangent function, and Fig. 7(b) is a graph after shifting to [0, 1]. The range of the inverse hyperbolic tangent function in the figure lies between -1 < x < 1, as shown in Fig. 7(a). Take one of the sections and shift it to the range of [0,1]. The range of the inverse hyperbolic tangent function after displacement is 0< x <1, so as to approximate the response curve of the human eye to Luminance. The inverse curve compensates for the lack of human visual response (the area in which human vision is insensitive), as shown in Figure 7(b).

Next, an example is given of how to define an adaptive inverse hyperbolic function using the adaptive deviation function bias ( x ) and the adaptive gain function gain ( x ). Please refer to Figure 8, which is a graph of various adaptive inverse hyperbolic tangent functions. Fig. 8(a) is a graph in which the adaptive gain parameter ( gain ) is fixed and the adaptive bias parameter ( bias ) is changed, and Fig. 8(b) shows that the adaptive deviation parameter ( bias ) is fixed and the adaptive gain parameter is changed ( Gain ). In Figure 8, we can see that the adaptive bias parameter ( bias ) helps us determine the turning point of the inverse hyperbolic tangent function and the shape of the curve. If the mean value ( mean ) of the luminance data is approximately equal to 0.5, the shape of the curve of the inverse hyperbolic tangent function is approximately a straight line, and the mapped pixel value approximates the original value. If the average value of the luminance data is less than 0.5, the shape of the curve of the inverse hyperbolic tangent function is mapped to a higher value. If the average value of the luminance data is greater than 0.5, the curve change of the inverse hyperbolic tangent function is mapped to a lower value. The definition of the bias ( x ) function is as follows:

The adaptive gain parameter ( gain ) determines the steepness of the curve. A curve with a larger slope maps a smaller input value to the display range of a larger output. Figure 8 (b) shows the mapping curves for the different gain values obtained for the different curves. The gain ( x ) function is defined as follows:

among them

Please refer to Figure 9 and Figure 10. Figure 9 is a comparison of the corresponding curves of the gamma function and the adaptive inverse hyperbolic tangent function. Figure 10 is the adaptive deviation bias parameter fixed to each value, adaptive gain gain A simulation of various state curves under parameter changes. The corresponding curve of the embodiment may include a corresponding curve of the cover gamma function, and the corresponding conversion will adapt to more varied images, and the effect is quite satisfactory. In other words, in the embodiment, the original brightness data of the input image is used to analyze the spectral state of the input image, and a contrast adjustment curve suitable for the input image is defined, and as shown in FIG. 10, corresponding to different brightness data. Curve. Thereby, the present embodiment can effectively improve image contrast and preserve the details of the original image brightness distribution.

On the other hand, please refer to FIG. 11 which is a flow chart of the steps of another embodiment of the present disclosure. In Fig. 11, in order to achieve a visual response curve conforming to human brightness and effectively improve image contrast, and preserve the details of the original image brightness distribution, the image with excessive brightness distribution of the image can be better compensated after processing; The method then develops into a multi-scale adaptive anti-hyperbolic tangent function, and adds the region's adaptive deviation function bias ( x ) and the region's adaptive gain function gain ( x ), which makes the brightness distribution of the image too extreme. The image is processed to get better compensation and improve image quality. Specifically, the present embodiment replaces the step 240 of the fourth FIG. 4 with the multi-scale adaptive inverse hyperbolic tangent function shown in step 270.

Please refer to FIG. 12A-D, and FIG. 12A-D is a detailed step flow chart of step 270 of FIG. In the 12A-D diagram, the present embodiment is based on the technique of the above-mentioned FIG. 4, and the processing technique of multi-scale is added as a reinforcement function. The proposed enhancement function is a multi-scale adaptive adjustment anti-hyperbolic tangent function to determine the radiance of each pixel. After reading the image file, we calculate the individual adaptive bias bias and the adaptive gain gain. Use as multi-scale parameter values and use these parameters to control the shape of the adaptive anti-hyperbolic curve. The multi-scale parameter is calculated from the original luminance data by the frequency band, and then the adaptive deviation bias and the adaptive gain gain parameter value are calculated for the luminance data of the divided frequency bands. Taking Figure 12A as an example, it is a two-scale image enhancement technique; firstly, the original luminance data is obtained from the input image, and the original luminance data is divided into two bands of high frequency band (H Band) and low band (L Band). . The adaptive deviation bias and the adaptive gain gain parameter values are calculated for the luminance data of the two frequency bands to generate a two-scale parameter, that is, a first-region adaptive deviation parameter ( bias ) and an adaptive gain parameter ( gain ) ( The Level 1 parameters shown in the figure are related to the second zone adaptive bias parameter ( bias ) and the adaptive gain parameter ( gain ) (Level 2 parameters shown in the figure). Further, respective adaptive inverse hyperbolic functions are defined to perform contrast adjustment. Finally, the adjusted adaptive brightness data is integrated with the original chroma data to generate an output image. Figure 12B is a four-scale image enhancement technique. The original luminance data is cut into high-height (HH Band), high-low (HL Band), low-high (LH Band) and low-low (LL Band) bands. A total of four bands of brightness data. Similarly, Figure 12C is an eight-scale image enhancement technique that operates in the same manner as described above. And so on, can be expanded to any N scale.

It is worth noting that the N- scale image enhancement technique depicted in Figure 12D is taken as an example; the multi-scale parameter method has two important design goals: 1. Avoiding the visibility of noise, especially in smooth areas, and 2 Prevent maximum and minimum possible luminance values from saturating (for example: 1 channel for each channel source format with values ranging from 0 to 255). The image output result of the multi-scale enhancement method is to process the input image x , which is described as follows:

Where k is the number of bands to be divided, and x _k is the luminance data of each sub-band (sub_band) after the band.

The above multi-scale technique is used to separate high brightness and low brightness levels. It automatically adjusts the local brightness of the image's high brightness and low brightness levels. Another feature is the ability to solve image types (such as backlit images) that cannot be processed by widely used gamma function curves. This conversion function has a higher rejection rate for higher luminance ranges and a higher elongation for low luminance regions. Therefore, it can be used not only for low dynamic range images, but also for high dynamic range images.

In summary, the adaptive anti-hyperbolic image processing method of the present disclosure has two main characteristics: (1) propose a multi-scale processing method to achieve regional contrast enhancement; (2) ) Propose a method that can handle extreme images and enhance or preserve the details of the original image. Combining the above two characteristics, and from the experimental results, we can confirm that the new multi-scale image contrast enhancement can improve the contrast enhancement of the image contrast enhancement technology and the contrast enhancement of the adaptive gamma function and the adaptive anti-hyperbolic tangent function image. It conforms to the human visual response curve of brightness.

In another aspect, the present disclosure provides an adaptive anti-hyperbolic image processing system including an input unit, an image processing unit, and an output unit. The input unit is configured to acquire an image; the image processing unit is configured to receive the image, and then perform the adaptive inverse hyperbolic image processing method; and the output unit is configured to output the processed output image.

Among them, the contrast type of the original image is determined by using new judgment criteria, and the conversion parameters are adaptively adjusted according to different contrast types, so the zoom space of the parameters is quite wide. The embodiment not only maintains the distribution shape feature of the histogram statistics of the original brightness and highlights the image details, but also effectively improves the contrast quality of the image. In other words, the present embodiment can adjust the contrast by using different corresponding curves for different brightness distributions, so as to adapt to the feelings of the human eye, and achieve the adaptability.

Finally, please refer to FIG. 1 and FIG. 2, which are images taken under light and insufficient conditions at dawn and night, respectively, through the conventional histogram equalization processing method and the image processing of the above embodiments. The comparative example after the contrast treatment was performed. Figure 1 and Figure 2(A) show the original image. (B) The plan is the image processed by the traditional histogram equalization. The two are distorted, and the filming time cannot be judged intuitively by its intuition. In other words, the traditional method is seriously distorted in brightness contrast. (C) is the image processing method according to the above fourth figure, using the adaptive anti-hyperbolic curve as the contrast adjustment curve, and the corrected image; the comparison (B) is closer to the brightness of the (A) picture, but still Without losing the clarity of its detailed performance. (D) The plan is the image processing method according to the above 11th image, and the multi-scale adaptive anti-hyperbolic curve is used as the contrast adjustment curve, and the corrected image; the image contrast and the detail processing are better than the (C) picture. However, due to the multi-scale processing procedure, the brightness data of the image is processed according to the spectrum, so the hardware resources are occupied.

The present invention has been disclosed in the above embodiments, and is not intended to limit the present invention. Any one of ordinary skill in the art to which the present invention pertains may make various changes and modifications without departing from the spirit and scope of the invention. The scope of protection of the present invention is therefore defined by the scope of the appended claims.

101-104. . . step

210-270. . . step

The above and other objects, features, advantages and embodiments of the present invention will become more apparent and understood.

Figure 1 is a luminance spectrum of various types of input images.

Figure 2 is a plot of the best fit contrast adjustment for the human eye corresponding to the various types of input images of Figure 1.

Figure 3 is a flow chart showing the steps of an embodiment of the present disclosure.

Figure 4 is a flow chart showing the steps of another embodiment of the present disclosure.

Figure 5 is a flow chart showing the detailed steps of step 204 of Figure 4.

Figure 6 is a graph of the processing of various types of images that the adaptive anti-hyperbolic tangent function can correspond to.

Figure 7 is a displacement plot of the adaptive inverse hyperbolic tangent function.

Figure 8 is a graph corresponding to the parameter values of the adaptive anti-hyperbolic tangent function.

Figure 9 is a comparison of the corresponding curves of the gamma function and the adaptive inverse hyperbolic tangent function.

Fig. 10 is a simulation diagram of various state curves under which the adaptive deviation bias parameter is fixed to each value and the adaptive gain gain parameter is changed.

Figure 11 is a flow chart showing the steps of another embodiment of the present disclosure.

Figure 12A-D is a detailed flow chart of step 270 of Figure 11.

101-104. . . step

Claims (6)

An adaptive anti-hyperbolic image processing method, comprising: step a: taking an original luminance data from an input image; and step b: setting an adaptive inverse hyperbolic function according to an average value and a standard deviation of the original luminance data; Step c: correcting the original luminance data by using the adaptive inverse hyperbolic function to generate an adaptive luminance data; and step d: modifying the input image by using the adaptive luminance data to generate an output image; wherein step b is The original luminance data is divided into a plurality of frequency band luminance data, and multi-scale parameters are generated by the average value and the standard deviation of the luminance data of the respective frequency bands, and a plurality of adaptive anti-hyperbolic functions are correspondingly set to one by one. Brightness data. The adaptive anti-hyperbolic image processing method according to claim 1, wherein the step a converts the three primary color data formats (RGB) of the input image into a chroma, saturation, and luminance data format. The adaptive inverse hyperbolic image processing method of claim 2, wherein the step d replaces the original luminance data with the adaptive luminance data. The adaptive anti-hyperbolic image processing method according to claim 1, wherein the step b is to calculate an adaptive deviation parameter according to an average value of the brightness data of each of the frequency bands, and according to the standard deviation of the brightness data of each of the frequency bands. Calculating an adaptive gain parameter, and then using the adaptive deviation parameter and the adaptive gain parameter to set the adaptive inverse hyperbolic function to correspond to the band luminance data. The adaptive anti-hyperbolic image processing method according to claim 1, wherein the step b calculates an adaptive deviation parameter according to an average value of the original brightness data, and calculates an adaptive gain parameter according to the standard deviation of the original brightness data. And using the adaptive deviation parameter and the adaptive gain parameter to set the adaptive inverse hyperbolic function. An adaptive anti-hyperbolic image processing system includes: an input unit for acquiring an image; and an image processing unit for receiving the image, thereby performing the method as claimed in claim 1, 2, 3 or 5. An adaptive anti-hyperbolic image processing method; and an output unit for outputting the output image.