On preprocessing technology of image recognition

In image recognition, the quality of image directly affects the design and effect accuracy of recognition algorithm, so in addition to the optimization of algorithm, preprocessing technology occupies an important factor in the whole project, however, people often neglect this point.

Image preprocessing, each text image is divided into a recognition module recognition, this process is called image preprocessing.

The main purpose of image preprocessing is to eliminate extraneous information in the image to restore useful real information to enhance the detection of information and to minimize the data so as to improve the reliability of feature extraction, image segmentation, matching and recognition. The preprocessing process generally has the steps of digitization, geometric transformation, normalization, smoothing, restoration and enhancement.

Digitizing the grayscale value of an original photo is a continuous function of the spatial variable (the continuous value of the position). In MX NThe image is sampled and quantified on the dot matrix (classified as one of the 2b grayscale levels), and the digital images can be processed by the computer. In order to reconstruct the original image, the size of M, n, and B values is required for the digital image to be reconstructed. The larger the values of M, N, and B are in the space and grayscale resolution range of the receiving device, the better the reconstructed image quality. When the sampling period is equal to or less than half the minimum detail period in the original image, the reconstructed image's spectrum is equal to the spectrum of the original image, so the reconstructed image can be identical to the original image. Because the product of M, N and B determines the storage capacity of an image in a computer, it is necessary to choose the appropriate m, N and b values according to the different properties of the image in order to obtain the best processing effect. The geometric transformation is used to correct the system error of the image acquisition system and the random error of the instrument position. The distortion caused by the system error of satellite image, such as Earth rotation, scanning mirror speed and map projection, can be expressed by the model and eliminated by geometric transformation. Random errors, such as aircraft attitude and height changes caused by the error, it is difficult to use the model to express, so generally in the system error is corrected, by the observed figure and the known correct geometric position of the graph compared, with a certain number of ground control points in the graph to solve the two-variable polynomial function group to achieve the purpose of transformation. Normalization is an image standard form in which certain features of an image have invariant properties under a given transformation. Some properties of an image, such as the area and perimeter of an object, are inherently invariant to coordinate rotation. Under normal circumstances, some factors or transformations on the image of some properties of the effect can be eliminated or weakened by normalization, which can be selected as the basis for the measurement of images. For example, for remote sensing images with uncontrolled illumination, normalization of gray histogram is necessary for image analysis. Gray normalization, geometric normalization and transformation normalization are the three normalization methods for the invariant properties of images. Techniques for smoothing out random noise in images. The basic requirement for smoothing technology is to eliminate noise without blurring the contour or line of the image. The common smoothing methods are median method, local averaging method and K-nearest neighbor averaging method. The size of the local area can be fixed, or it can vary by point by the size of the grayscale value. In addition, the spatial frequency domain Bandpass filtering method is sometimes applied. Restoration corrects image degradation caused by various causes, so that reconstructed or estimated images are approximated as close as possible to the ideal non-degenerate image field. Image degradation is often occurring in practical applications. For example, the disturbance of the large airflow, the aberration of the optical system, the relative motion of the camera and the object will degrade the remote sensing image. The basic restoration technique is to treat the acquired degraded image g (x, y) as a convolution of the degenerate function h (x, y) and the ideal image f (x, y). Their Fourier transforms exist in relation to G (U,v=h (u,v) F (u,v). After determining the degenerate function according to the degradation mechanism, F (u,v) can be obtained from this relation, and then the Fourier inverse transform is used to find the F (X, y). Usually referred to as a reverse filter. In practical application, as H (u,v) decreases rapidly with the distance from the origin of the UV plane, in order to avoid the enhancement of noise in the high frequency range, when the U2+V2 is greater than a certain limit value w 娿, M (u,v) equals 1. The choice of W0 should cause H (u,v) not to appear 0 points in the U2+v2≤w 娿 range. The algebraic method of image restoration is based on the best criterion of least squares. Seeking a valuation assist to minimize the value of the merit criterion function. This method is relatively simple and can deduce the least squares Wiener filter. When there is no noise, the Wiener filter becomes the ideal reverse filter. Enhanced selective enhancement and suppression of the information in the image to improve the visual effect of the image, or to transform the image into a more suitable form for machine processing for data extraction or identification. An image enhancement system, for example, can highlight the contour of an image through a high-pass filter, allowing the machine to measure the shape and perimeter of the contour line. There are several methods for image enhancement, such as contrast broadening, logarithmic transformation, density stratification and histogram equalization, which can be used to change the image gray and highlight details. In practical application, it is often necessary to use different methods to test repeatedly to achieve satisfactory results.

On preprocessing technology of image recognition