# CS代写|图像处理作业代写Image Processing代考|GPY470

## CS代写|图像处理作业代写Image Processing代考|Classify Noise Filtering Results in Seismic Images

Seismic imaging is the main technology used for subsurface hydrocarbon prospection. The raw seismic data is heavily contaminated with noise and unwanted reflections that need to be removed before further processing. Therefore, the noise attenuation is done at an early stage and often while acquiring the data. Remote sensing images are often affected by noise in the process of digitization and transmission processes. De-noising is an indispensable way of improving image quality (Wu et al. 2020).

To give confidence in the de-noising process, quality control $(\mathrm{QC})$ is required. It can ensure that a costly data re-acquisition is not needed. A supervised learning approach to build an automatic QC system is proposed (Mejri and Bekara 2020). The QC system is an attribute-based classifier that is trained to classify three types of filtering (optimal = good filtering; harsh = over filtering, the signal is distorted; mild = under filtering, noise remains in the data). The attributes are computed from the data and represented geo-physically for the statistical measures of the quality of the filtering.

Experimental results show that some attributes show a good level of visual separation between the different types of filtering, particularly the harsh one. The clusters of attributes for the mild and the optimal filtering are close as they reflect the observation made earlier about the subtle differences between the two types of filtering.

In this system, multi-layer perceptron (MLP) is used for classification. Based on the fundamental assumption that without regularization, an optimal MLP structure tends to over fit as its depth is increased, a secure MLP-building strategy is adopted to build the multilayer perceptron model with the least validation cross-entropy error.

Figure $2.27$ depicts the flowchart of the one-vs-all binary classification process. The training and the validation sets were split into $K$ subsets of equal size using the bootstrap aggregation strategy. The training and validation subsets were converted to $N=3$ binary sets. Then, the binary MLP generator is used to predict binary decision subspaces. The predicted class members are then fed forward to a voting system, which decides on the class based on the majority vote, resulting in more accurate classification results.

Image blur is a common image degradation situation or process, which can also be represented by the model shown in Figure 3.1, where the blur is generated by the system $H$, which is also called the blur kernel at this time. In Figure 3.1, the input image is affected by both blur and noise. As a result, the degraded image $g(x, y)$ becomes a noisy blurred image.
If the blur kernel is known, only the clear image needs to be solved. This problem can be called non-blind de-blurring. If both the blur kernel and the clear image need to be solved, this problem is called blind de-blurring. For the blind de-blurring problem, if the blur kernel can be obtained, the problem can be transformed into a non-blind de-blurring problem.
Solving the blind de-blurring problem directly is an ill-conditioned (underdetermined) problem because the solution is not unique. In the actual solving process, it is necessary to introduce prior knowledge about the blur kernel or clear image (including the heuristic knowledge of enhancing the edge and the knowledge of constructing the prior probability distribution model). According to the mathematical method to solve the underdetermined problem, one can consider constructing a regularized cost function based on the prior information of the image to transform the problem into a variational problem, in which the variational integral depends on the data and smoothing constraints at the same time. For example, for the problem of estimating function $f$ from a set of values $y_1, y_2, \ldots, y_n$ at points $\boldsymbol{x}1, \boldsymbol{x}_2, \ldots, \boldsymbol{x}_n$, the regularization method is to minimize the functional $$H(f)=\sum{i=1}^N\left[f\left(x_i\right)-y_i\right]^2+k \Phi(f)$$
In the equation, $\Phi(f)$ is a smoothing functional, and $k$ is a positive parameter, which is called a regularization number.

The blur caused by different reasons will have different effects on the image quality (different changes in the image), and the blur kernels corresponding to different types of blur can also be very different.

The necessary condition for motion blur is that there is relative motion between the camera and the object during the imaging process. This motion can originate from camera motion (global motion), object motion (local motion), or both. In practice, if the imaging time is long and/or the motion is relatively violent, resulting in the length of the trajectory of the motion reaching the pixel level during the imaging process, visible motion blur will be formed on the captured image. Motion blur is embodied in the image that the scenery stretches along the direction of motion and produces double shadows, so it is also called motion smear. The image acquisition system (with a narrow field of view) that uses a telescope lens is very sensitive to this type of image degradation.

The out-of-focus blur is related to the depth of field of the camera lens. The depth of field of the lens corresponds to the distance between the closest object and the farthest object that can be clearly imaged in the scene. When the camera lens is focused at a certain distance, the scene at that distance is the clearest, and the scene deviating from this distance will gradually blur with the degree of deviation. In general, within a certain range (depth of field) before and after this distance, the blur does not reach the pixel level, and the resulting blur cannot be noticed; the scene beyond this range will show the blur effect on the collected image. This blur is generally isotropic, which limits the resolution sharpness of image. Therefore, if it does not focus on the object one wants to observe (missing focus), the object in the image may not be clear enough.

# 图像处理代考

## CS代写|图像处理作业代写Image Processing代考|Classify Noise Filtering Results in Seismic Images

H(F)=∑一世=1ñ[F(X一世)−是一世]2+ķ披(F)

myassignments-help数学代考价格说明

1、客户需提供物理代考的网址，相关账户，以及课程名称，Textbook等相关资料~客服会根据作业数量和持续时间给您定价~使收费透明，让您清楚的知道您的钱花在什么地方。

2、数学代写一般每篇报价约为600—1000rmb，费用根据持续时间、周作业量、成绩要求有所浮动(持续时间越长约便宜、周作业量越多约贵、成绩要求越高越贵)，报价后价格觉得合适，可以先付一周的款，我们帮你试做，满意后再继续，遇到Fail全额退款。

3、myassignments-help公司所有MATH作业代写服务支持付半款，全款，周付款，周付款一方面方便大家查阅自己的分数，一方面也方便大家资金周转，注意:每周固定周一时先预付下周的定金，不付定金不予继续做。物理代写一次性付清打9.5折。

Math作业代写、数学代写常见问题

myassignments-help擅长领域包含但不是全部: