A Center Symmetric Padding Method for Image Filtering
Padding the boundary is the first step in image filtering. If not appropriately handled, it often cause serious artifacts and losing information. In this paper an optimized boundary padding method is proposed to solve this problem. We analyze the pixel va
- PDF / 561,234 Bytes
- 7 Pages / 439.37 x 666.14 pts Page_size
- 60 Downloads / 193 Views
)
College of Mechatronics and Control Engineering, Shenzhen University, Shenzhen, China [email protected], [email protected]
Abstract. Padding the boundary is the first step in image filtering. If not appro‐ priately handled, it often cause serious artifacts and losing information. In this paper an optimized boundary padding method is proposed to solve this problem. We analyze the pixel values near the boundary, calculate the gradient of the pixels, and pad the input image with a center symmetric padding method. Consequently, the details of the boundary will be preserved, and the number of unexpected noise will be suppressed as much as possible. We demonstrate that the center symmetric padding method is more effective than the traditional border handling methods in some computer vision applications.
1
Introduction
In computer vision, most of applications have to do some preprocessing to reduce noise or get useful information. Image filtering is a general and effective method in image processing such as non-linear filter in [1] and most of linear filter. Image filtering is generally modeled as a convolution of a pixel’s neighborhood. The convolution operator makes use of not only the image in the Field of View (FOV) of the given observation, but also part of the scenery in the area bordering it. The missing pixel information outside the observed image can be synthesized by extrapolating the available image data. In this paper, we propose a boundary padding application when dealing with the flaw detection of the backlight plate in cell phone. The picture of the backlight plate is taken from the line-scan CCD camera, whose resolution is 8192*14000, see in Fig. 3. In this case, padding the boundary which is the first step in image filtering plays an important role in such an extremely high resolution. Border handling provides approx‐ imate output pixels close to the boundary of the image where insufficient data is available to fully compute the local operator. There are many available methods to padding the boundary. The “tile” method in [2] proposed an algorithm that pads the image by repli‐ cating small tiles of pixels adjacent to the border to reduce boundary artifacts in image deconvolution. Halide*’s [3, 4] library also includes “tile”. Tile padding is particularly useful for Bayer coded color images because it preserves the Bayer color pattern. Most of functions in commercial software or open source library are copying a given image onto another slightly larger image and then automatically padding the boundary, such as Reflect, Reflect101 [5], Replicate [6], Wrap, Constant. But these boundary methods not always work well, especially when the blur filter window size is relatively larger with a high resolution of the backlight plate.
© Springer Nature Singapore Pte Ltd. 2017 K. Kim and N. Joukov (eds.), Information Science and Applications 2017, Lecture Notes in Electrical Engineering 424, DOI 10.1007/978-981-10-4154-9_15
A Center Symmetric Padding Method for Image Filtering
125
In this paper, we proposed a simple but effe
Data Loading...
