CN108389173B

CN108389173B - Opportunity cost-based parameter optimization method

Info

Publication number: CN108389173B
Application number: CN201810248394.0A
Authority: CN
Inventors: 朱思霖; 杨鹏
Original assignee: Xiamen University Tan Kah Kee College
Current assignee: Xiamen University Tan Kah Kee College
Priority date: 2018-03-24
Filing date: 2018-03-24
Publication date: 2021-08-31
Anticipated expiration: 2038-03-24
Also published as: CN108389173A

Abstract

The invention relates to a parameter optimization method based on opportunity cost, which comprises the steps of firstly inputting n images, then calculating PSNR values of different images under all parameter combinations, calculating the opportunity cost value of each image, then calculating the opportunity cost total value of all images, and finally selecting the parameter combination with the minimum opportunity cost total value as an optimal parameter combination. The method introduces the opportunity cost thought in economics to the image enhancement field and selects the optimal parameter combination.

Description

Opportunity cost-based parameter optimization method

Technical Field

The invention relates to the field of nonlinear enhancement of images, in particular to a parameter optimization method based on opportunity cost.

Background

The opportunity cost means that when a certain choice is made, the analysis process of other possible choices has to be abandoned. When a decision is forced to be made to take either of the two, there is a cost of opportunity that is the value of the best available decision for the other. Opportunistic cost analysis methods focus on the cost of the analysis selection. In life, some opportunity costs can be measured in currency. For example, farmers may not choose to raise their chickens if they choose to raise their pigs when they have more land available, and the opportunity cost of raising their pigs is to give up the benefits of raising their chickens. But some opportunity costs are often not measured in currency, for example, to choose between reading books in a library and enjoying the enjoyment of a television show.

Opportunity cost generally refers to the greatest loss of the choice, and the opportunity cost changes with the cost of the choice, for example, when the preference or value of the rejected choice changes, the value obtained does not change. The following example is taken as a calculation example: when farmers obtain more land, if the selection of pig raising is carried out, other poultry can not be selected to be raised, and the opportunity cost of pig raising is the income of abandoning chicken raising or duck raising and the like. If pig raising can obtain 9 ten thousand yuan, chicken raising can obtain 7 ten thousand yuan, and duck raising can obtain 8 ten thousand yuan, the opportunity cost of pig raising is 8 ten thousand yuan, the opportunity cost of chicken raising is 9 ten thousand yuan, and the opportunity cost of duck raising is 9 ten thousand yuan.

In the field of information technology, for an algorithm with non-single parameters (the number of parameters is two or more), the selection of various parameter values is often a trade-off. The cost of selecting the optimal parameters is to forego the potential benefits of suboptimal selection. The theory is often used in economics and related areas. In the field of information technology, algorithms or architectures related to each link are often multi-parameter or multi-technical indexes. If an image is to be obtained after the original image is enhanced, the image G to be enhanced can be obtained by adding the corresponding high frequency component to the low frequency component of the image in the field of nonlinear enhancement of the image₀If it is regarded as a low-frequency image, a corresponding higher-frequency image L can be constructed_-1By superposition of G₀And L_-1Will obtain the ratio G₀Clearer image G_-1(i.e., the enhanced image) the basic technical flow is as shown in the attached figure 1 of the specification. For G₀Middle high frequency component L₀Performing nonlinear filtering, and extrapolating₀Phase-aligned higher frequency image L_-1Thereby realizing the enhancement of high-frequency components of the blurred image.

For G₀Middle high frequency element L₀Performing secondary derivation and quartic derivation to obtain L₀And L₀", so by adjusting L₀And L₀"" then, a more approximate ideal high-frequency component L can be obtained by the formula (3-1)_-1Is/are as follows

Wherein k is₁And k is₂The method can be used to generate the high-frequency component required for amplitude regulation

Then, the obtained high frequency component can be obtained

And an input image G₀Adding to obtain enhanced result image

Namely:

when the two-dimensional image is subjected to enhancement processing, a classical linear Laplacian high-pass filter can be used for carrying out high-frequency component L₀And (5) calculating a second derivative and a fourth derivative. In the experimental simulation, the filter can be replaced by a 3 × 3Laplacian template shown in the attached fig. 2 of the specification.

In this image enhancement algorithm, the parameter k₁And k is₂The value of (a) will have an effect on the final enhancement effect. In the case of objectively evaluating image quality, the quality of the image is actually reflected by the size of an objective parameter related to the image, and here, a Peak Signal Noise Ratio (PSNR) value is used for description. Generally, the greater the PSNR value, the better the image will be when comparing different enhancement algorithms. The PSNR is specifically defined as follows:

wherein 2ⁿ-1 is the maximum possible pixel value in the image, n is the number of bits required for each pixel, and for an 8bit representation of the image this value is 255.

As can be seen from the formula (3-3), if g_m,n≈f_m,nThe PSNR value approaches infinity, and the distortion caused by the processing is small and the image quality is good.

Disclosure of Invention

In view of this, the present invention provides a method for optimizing parameters based on opportunity cost, which introduces the opportunity cost thought in economics into the image enhancement field and selects an optimal parameter combination.

The invention is realized by adopting the following scheme: a parameter optimization method based on opportunity cost specifically comprises the following steps:

step S1: given n input images, setting a parameter affecting high frequency components of the images to be k₁And k is₂(ii) a For each image, each group k₁And k is₂Can generate corresponding PSNR values by setting m groups of different k₁And k is₂A combination of parameters of (1);

step S2: calculate all k₁And k is₂Calculating the PSNR (i, j) of different images under the combination of (a) and (b); wherein i represents k₁And k is₂J represents the serial number of the current image, i is more than or equal to 1 and less than or equal to m, and j is more than or equal to 1 and less than or equal to n;

step S3: calculating the maximum PSNR value of each image, wherein the maximum PSNR value Optimal _ PSNR (j) of the jth image is calculated by adopting the following formula:

Optimal_PSNR(j)＝Max(PSNR(1:m,j))；

step S4: calculating an opportunity cost value of each image, wherein the opportunity cost value of the jth image cost (j) is calculated by the following formula:

Cost(j)＝|Optimal_PSNR(j)-PSNR(i,j)|；

step S5: the total opportunity cost value Tcost for all images is calculated using the following equation:

step S6: selecting the i value corresponding to the minimum Tcost and the k corresponding to the i value₁And k is₂Combine and select k₁And k is₂The combination is the most optimal parameter combination.

Further, in step S1, m ≦ n.

Compared with the prior art, the invention has the following beneficial effects:

1. the method applies the opportunity cost in economics to the field of image enhancement, and can obtain optimal parameters in a certain sense for a multi-parameter algorithm.

2. The algorithm of the invention can realize the automatic initialization selection process of the parameters for the multi-parameter algorithm, and finally can finely adjust and optimize the parameters according to the specific algorithm.

Drawings

Fig. 1 is a basic block diagram of a non-linear image enhancement scheme in the background of the invention.

Fig. 2 is a schematic diagram of a linear high-pass Laplacian filter template M in the background art of the present invention.

Fig. 3 is a schematic diagram of the algorithm in the embodiment of the present invention.

FIG. 4 is a schematic diagram of a test image according to an embodiment of the present invention.

FIG. 5 is a schematic diagram of an application of the embodiment of the present invention.

Detailed Description

The invention is further explained below with reference to the drawings and the embodiments.

The present embodiment provides a method for parameter optimization based on opportunity cost, which is the loss caused by selecting one solution among a plurality of different solutions to meet a specific objective. For example, let X₁And X₂Respectively representing two images, selected for X₂Will generate the opportunity cost (X) corresponding to it₂₂) In the case of this scheme, the result is also X₁Opportunity cost (X)₂₁). Likewise, select for X₁Will generate the opportunity cost (X) corresponding to it₁₁) In the case of this scheme, the result is also X₂Opportunity cost (X)₁₂). Adding the opportunity costs of different images for the same solution can result in X for image₁And X₂Total opportunity cost in terms of, i.e. total opportunity cost Tcost of solution 1₁And the total opportunity cost Tcost of solution 2₂Comparing their sizes, selecting a group of smaller values to obtain the image X for enhancement₁And X₂In terms of an optimal solution based on opportunity cost. In the above example, the opportunity cost (X)_ij) The subscripts in (a) are defined as: the first index i indicates the optimal solution for the image and the second index j indicates the corresponding image under the solution.

Given n (n > 1) input images, for each image, each group k₁And k is₂The parameter combination value of (2) can generate corresponding PSNR values, and the optimal parameter combination value corresponding to the maximum PSNR value can be easily found out from the PSNR values. Assuming that there are m sets of parameter combination values, it is obvious that m ≦ n, since the optimal parameter combination is the same for some images. In the m sets of parameter combinations, each image will have m opportunity cost values, the opportunity costs of different images in each set of parameter combinations are added to obtain m total opportunity cost values Tcost, the corresponding parameter combination value when the minimum value of Tcost is selected is the optimal parameter combination value, and the specific example of the algorithm is shown in fig. 3.

The method specifically comprises the following steps:

Optimal_PSNR(j)＝Max(PSNR(1:m,j))；

Cost(j)＝|Optimal_PSNR(j)-PSNR(i,j)|；

In the present embodiment, m ≦ n in step S1.

Since the optimal parameter combination is difficult to obtain intuitively, we define k empirically₁And k is₂The parameter value range of (2). In simulation experiments, k₁And k is₂Respectively has a value of k₁0.05,0.10,0.15,. 1 and k₂0.1,0.2, 0.3. 12 standard gray scale images with the size of 512 x 512 are selected, as shown in fig. 4.

TABLE 1 PSNR (dB) value analysis under different parameters of different images

TABLE 2 different k₁、k₂Opportunity cost and total cost corresponding to parameter combination

X1 to X2 in Table 1 and Table 2 correspond to images Aeral, Airplane, Bike, Couple, Crowd, FlintStones, Elaine, Lena, Parrot, Peppers, Stream and bridge, and Tank, respectively. Table 1 shows the PSNR values of the 12 images under the respective optimal parameter combination valuesSo as to see that the total number of the 12 selected images has 3 groups of optimal parameter combination value options, which are respectively k₁0.05 and k₂＝0.3、k₁0.05 and k₂0.4 and k₁0.05 and k₂0.5. Table 2 shows the differences of k₁And k is₂The opportunity cost and the total cost corresponding to the optimal parameter combination can be obviously seen₁0.05 and k₂The total opportunity cost value Tcost corresponding to 0.4 is the minimum, so the set of parameter combinations is optimal.

TABLE 3 PSNR (dB) Performance comparison of the present enhancement algorithm

For the algorithm of the present embodiment, when using the opportunistic cost parameter optimization algorithm, it can be seen from table 3 that k is₁＝0.05,k₂0.4 is the optimum parameter combination for all test images, and the total opportunity cost value Tcost corresponding to the combination is not only the minimum (see table 2), but also the PSNR value for the test images is excellent.

The embodiment can be applied to the field of nonlinear enhancement algorithms, as shown in fig. 5.

The above description is only a preferred embodiment of the present invention, and all equivalent changes and modifications made in accordance with the claims of the present invention should be covered by the present invention.

Claims

1. A parameter optimization method based on opportunity cost is characterized in that: the method comprises the following steps:

Optimal_PSNR(j)＝Max(PSNR(1:m,j))；

Cost(j)＝|Optimal_PSNR(j)-PSNR(i,j)|；

step S6: selecting the i value corresponding to the minimum Tcost and the k corresponding to the i value₁And k is₂Combine and select k₁And k is₂Combining the most optimal parameter combination;

in step S1, m is less than or equal to n.