Precision positioning is very important for automatic assembly and inspection in the electronic manufacturing process. In this paper, we propose a fast image alignment method using the expectation–maximization (E–M) technique. The proposed algorithm is especially applied to positioning and defect inspection of printed circuit boards (PCBs). It can well handle deformed or incomplete object shapes with translation, rotation, and scale changes. The Canny edge detector is used to generate the edge maps of images. The E-step of the E–M procedure finds mutual edge points in both compared images by assigning weights to individual edge points. The mutual edge points give larger weights, while the foreign edge points in two images have smaller weights. The M-step then calculates the geometric transformation parameters using the weighted edge points in individual images. For an edge point in one image, a fast spiral search is proposed to find its corresponding edge point with the shortest distance in the other image. The spiral search is carried out by a predetermined lookup table, and no computation is involved in the search process. The weight of each edge point is inversely proportional to the neighboring distance. Experimental results indicate that the proposed E–M positioning method can achieve a translation error less than 1 pixel and a rotation error smaller than 1° for PCB positioning.

Machine Vision-Based Positioning and Inspection Using Expectation–Maximization Technique

Since their introduction, FPGAs can be seen in more and more different fields of applications. The key advantage is the combination of software-like flexibility with the performance otherwise common to hardware. Nevertheless, every application field introduces special requirements to the used computational architecture. This paper provides an overview of the different topics FPGAs have been used for in the last 15 years of research and why they have been chosen over other processing units like e.g. CPUs.

Survey of FPGA applications in the period 2000 – 2015 (Technical Report)

Image registration is a crucial step in some specific applications of remote sensing. High resolution remote sensing images have made it more convenient for people to observe the Earth; however, they also create challenges for traditional research methods. In terms of image registration, there are a number of problems with using conventional image registration techniques for high resolution images. This paper develops an improved feature-based geometric registration approach for high spatial resolution images, in which control points are efficiently selected automatically. First a modified watershed transformation algorithm is used for image segmentation. Then regions are represented by centers of gravity, and road intersection can be considered the main target to perform the control point extraction. Secondly the extracted points in the reference and warped images can be matched using spatial relations. Finally, pixels in the warped image are directly corrected using the chosen geometric model followed by the corresponding interpolation methods. The proposed feature-based geometric correction method is implemented using a time series of QuickBird images acquired over the 2004/2005 summer season at Xuzhou City, Jiangsu Province of China. The experimental results show that the presented rectification approach is convenient to determine ground control points, and is efficient for high spatial resolution image geometric registration.

Feature-based geometric registration of high spatial resolution satellite imagery

This paper describes a method against geometric distortion of digital watermarking image by means of Hausdorff distance measuring, which matches the feature points, and calculates similarity between two point sets (maximum distance). In this paper, Harris feature points are firstly extracted into feature point set according to the response of the Harris detector and the field size, and then secondly the Hausdorff distance is used to eliminate the image geometric deformation, thirdly, DWT(Discrete Wavelet Transform) transformation is conducted with the Harris feature points as centers, and also DCT(Discrete Cosine Transform) transformation is conducted in low-frequency LL domain; at last, watermark could be embedded and extracted in the DCT coefficient. The experimental results demonstrate that the embedded watermark image is well extracted and has high similarity after such geometric attacks as translation, rotation, scaling and shearing, so this method is a blind watermarking detection algorithm with strong robustness.

A method on digital watermarking image against geometric distortion

Item recognition has become a hotspot in the field of computer vision research. SIFT has the advantage of requiring a low amount of information, a fast running speed and high precision, but it requires large data calculations and thus takes a long time to perform the item recognition. In this paper we propose a method of item recognition based on SIFT and SURF that provides a new way to solve the problem of item recognition, and has both feasibility and availability. This technique currently ignores colour information when dealing with colour images, but the evaluation method is capable of taking colour quality characteristics into account so it should be possible to improve the algorithm in the future. Experimental results show that this system of item recognition based on the SURF algorithm gives better matching recognition, is faster and has greater robustness.

Method of item recognition based on SIFT and SURF

As for the RST transform in image registration, corresponding formula of box distance transform is deduced. Compared with traditional formula of general affine Hausdorff box distance, search range of distance is reduced. The paper proposes regional Voronoi surface combining comparison of sliding windows when computing Hausdorff distance, characterized by reducing calculating-cost for Voronoi surface. It also has the advantages of eliminating trivial edges and preserving longer edges for calculating. Experimental results show that calculation speed of image registration based on Huasdorff distance is improved.

Image registration based on Hausdorff distance

Raptor was valued and welcome in the field of broadcast and internet application, because it possesses properties of rateless, linear time with encoding and decoding, and applicability to erasure channel especially. For the most part of the applications, real time was generally required. Hereby, delay should be taken into consideration in the design and realization of a system. Channel encoding and decoding speed will also influence the size of delay finally, so the speed of Raptor encoding and decoding is worth to study and improve. This paper adopts several methods to optimize Raptor encoding and decoding in software realization, and we get a satisfactory result with improved speed of encoding and decoding.

Raptor algorithm optimization with software coding

Serial Digital Interface (SDI) has become a standard video interface gradually. The former methods to implement the SDI cause much costs and low agility. Aiming at the situation that analogue and digital coexist, this paper uses a FPGA-based way to design the SDI, it mainly expounds the scheme of the SD SDI design, and analyses its structures and function of each module, which is implemented by Hardware Description Language.

Conversion from CVBS to SDI based On FPGA

A new method based on the edge curvature area extreme-value and the edge correlation for efficient image registration is proposed. First of all, draw some edges from the pending image, set the edge length threshold by using a comparatively large Gauss parameter, save the longer edges only, and mode them with improved Freeman. Secondly, with the Freeman code aforementioned, compute the curvature of the curve, and match the edges according to the corresponding situation of the curve area curvature and get the scale transform parameter. Thirdly, resample the image by using that scale parameter, draw some edges again and by applying the edge correlation and consistency check means, get the corresponding pairs of edges. Finally, by using the Least Mean Square, obtain the registration parameter of the image. This arithmetic applies in situations like registration images with comparatively rotating angles and translational motions.

A multi-sensor image registration method adapted for larger scale

Some registration approaches can fail when percentage of outliers is too high in remote images. We introduce, in this paper, a new approach to improve the robustness of feature extration for automatic image registration. This method is based on long-edge correlation and consistency check. Long-edge correlation extracts a long edge as reference curve in order to increase the percentage of common feature in the edge maps, and consistent check reduce the number of outliers drastically. The proposed method based on comparison of HuiLi's correlation is a modified chain code correlation coefficient method. In addition, get more consistent-edge by improvement of Randrianarisoa method. The simulation experiments show the robust registration results of the method for images rich in long-edge. Another advantage of the method is rapid computational speed.

Niu Li-pi

Papers

Image registration based on Hausdorff distance

Raptor algorithm optimization with software coding

Conversion from CVBS to SDI based On FPGA

A multi-sensor image registration method adapted for larger scale

A multi-sensor image registration approach based on long-edge-correlation