When people can’t get close to identify the number of instruments, the digital area on the display screen can only be transmitted to the computer by the camera, and the digital recognition is realized through the computer’s related software and algorithms.After image pre-processing operations, such as geometric correction, gray-scale, threshold and corrosion expansion, the projection method is used to segment the image and the number is identified by the sewing method. Compared with the traditional method of digital recognition, the efficiency of the sewing method is very high. https://vssewingmachine.in/ The combination features of the horizontal and vertical thread of the image is used to complete the cascade classification after the character preprocessing operation. In the digital image acquisition and recognition system based on sewing method, 4320 numbers are identified and tested, of which an average of 122 digits can be handled per second. This method performs well in processing speed, recognition accuracy and anti-interference.
n order to judge whether the instrument produced by the company meets the industry standard, the instrument should be sent to the Metrology Institute for testing. The identification results of the weighing apparatus are compared with the technical requirements and standards of the industry to determine whether the weighing instrument meets the technical requirements and standards of the industry. [1] The process of digital acquisition is usually carried out in the case of high frequency electromagnetic interference, where people cannot get close to. The digital area on the display screen can only be captured by the camera and transmitted to the computer, and the digital recognition can be realized by the relevant software and algorithm of the computer. [2]The traditional digital recognition includes threshold, de-noising, feature extraction, contour search, template matching and other processing steps [3]. Their efficiency is very low, while their requirements for system hardware are very high. The addition of BP neural network and other algorithms will make the hardware and software more difficult and not suitable for real-time processing. Compared with the traditional method of digital recognition, the efficiency of the sewing method is very high [4]. This algorithm uses the combination features of the horizontal and vertical thread of the image after the character preprocessing operation to complete the cascade classification. Digital recognition of sewing method performs well in processing speed, recognition accuracy and anti-interference. However, because alphabet units are generally smaller than figures, so the number of letters cannot be identified by the sewing method [5
IMAGE RECOGNITION PROCESS
The whole image recognition process can be divided into image geometric correction, image preprocessing, digital segmentation and recognition. The camera captures the digital area of the instrument at first. In order to avoid the shadow of the camera affecting digital recognition, it is necessary to make a certain angle between the camera and the display screen, so that the captured image is trapezoidal. Figure 1 work flow of digital recognition At this time, the image is needed to be converted into a rectangle by the geometric correction, and then the images should be pretreated with some related preprocessing operation, such as threshold, gray scale, and corrosion expansion and so on. Then, the projection method is used to segment the numbers. Finally, the numbers are recognized by sewing method.
DIGITAL RECOGNITION ALGORITHM
Image geometric correction If the camera is perpendicular to the LCD or LED digital display, the shadow of the number will overlap on the display data of the digital instrument. VS Sewing Machines The problem can be solved when there is a certain angle deviation between the camera and the display screen [6]. A slant graph is shown in Figure 2, but the shape of the captured image is trapezoidal rather than rectangular, while trapezoid image will make false digital region segmentation. he work of this paper is to calibrate the image acquired by tilting, and then a rectangular image is generated. The basic process is divided into two steps: coordinate transformation and gray scale reconstruction. The coordinate transformation can be divided into the forward mapping method from the distorted image to the ideal image and the backward mapping method from the ideal image to the distorted image. Gray reconstruction includes interpolation, bilinear interpolation and cubic convolution interpolation. [7] The accuracy of adjacent interpolation method is not high, while cubic convolution interpolation method has high accuracy but large amount of computation. Bilinear interpolation can be combined with the advantages of the two algorithms, and is the most appropriate. In this paper, a transformation relation correction image from distorted image to ideal image is searched, and the gray level reconstruction is performed by bilinear interpolation.
(1) Gray-scale processing First of all, the gray scale processing of digital pictures is to transform RGB color images into gray images. The color information in the color image is removed and the brightness information is kept [8]. According to the relationship between RGB tricolor and YUV (brightness, chrominance, concentration), a formula is established: Y=0.3R+0.59G+0.11B. The information contained in gray-scale image is much smaller than that in color image, which simplifies the process and does not affect the accuracy of recognition. It is the basis for the steps of the subsequent image processing. (2) The threshold processing The picture is transformed into only two gradations, that is, only black and white in the image. The images produced by this method are smaller and more convenient for subsequent image processing [9]. The adaptive threshold is used as a variable, which is different at each pixel point. By calculating the weighted average of the area at the around pixels, an adaptive threshold can be obtained by subtracting a constant. (3) Morphological processing There are still some interference factors after threshold of the image. In order to eliminate these interference factors, we need to make morphological treatment of the picture, that is, expansion and corrosion treatment. These characters are connected by expansion treatment, and small fragments are eliminated by corrosion. Expansion or corrosion operations are the convolution of images and nuclei, and the core can be any shape and size. Expansion:( , ) max src(x+x',y+y') (x',y'): element(x',y') 0dst x yThe expansion is to seek the local maximum value, calculate the maximum value of the pixels in the core coverage area and assign it to the specified pixel of reference point. Corrosion is contrary to expansion. Corrosion:( , ) min src(x+x',y+y') (x',y'): element(x',y') 0
