Issue34

Z. Jijun et alii, Frattura ed Integrità Strutturale, 34 (2015) 590-598; DOI: 10.3221/IGF-ESIS.34.65 591 D ETECTION METHOD OF BROKEN WIRE ROPE Image preprocessing based on the improved homomorphic filtering optimization algorithm s shown in Fig. 1, the collected images all have relatively strong side light and serious degradation with uneven brightness since the experimental environment is very complex, bringing great difficulty to the following texture extraction. Thus, the images should be dealt with filtering processing to different degrees. Figure 1 : Original drawing of wire rope Analysis shows that the background information of the image corresponds to the high-frequency part of the image, while the information of the object itself corresponds to the low-frequency part. For the image with uneven brightness, we need to weaken the low frequency component and strengthen high frequency component to reflect features of the object to the greatest extent and strengthen contrast ratio, and then conduct homomorphic filtering [4]. The classical homomorphic filtering algorithm is conducted in the frequency domain. Firstly, FFT exchanges will be made to the image, and different filter functions will be adopted for filtering according to the needs of image’s the low-frequency part and high-frequency part. IFFT exchange will be conducted in the last period, and then turn the image back. There are several obvious disadvantages concerning the frequency domain algorithm. The first one is the failure to achieve the effect of local contrast enhancement; the second is that two times of FFT will reduce the efficiency of operation due to the big calculation amount of FFT. To solve these problems, Conducting homomorphic filtering on spatial domain is taken into consideration. Generally, it will firstly divide the original image into low-frequency part (incidence component) and high- frequency part (reflection component). Next, conduct filtering to the images with Gaussian Lowpass Filters to get the component of incident light. The original image subtracts the incident light to result in the reflection component. Since the incident component and reflection component are separated by taking logarithm, the true image with enhancement can be obtained with antilog operation. In addition, the thought of template factoring has improved the Gaussian Lowpass Filters and greatly reduced the space for operation and quickened the arithmetic speed, causing a satisfactory effect of homomorphic filtering. Consequently, the paper adopts spatial homomorphic filtering algorithm [5-6], to correct the brightness of the image of wire rope, so as to eliminate the influence of the uneven illumination. The detailed steps are as follows: ( , ) f x y , function of gray level of the image can be seen as the product of the incident light and reflected light, namely: ( , ) ( , ) ( , ) f x y i x y r x y   ( , ) i x y is incident light, and 0< ( , ) i x y <∞, while ( , ) r x y is the reflected light, and 0 ( , ) i x y    . Then take the logarithm of the image, that is assuming: ( , ) ln ( , ) ln ( , ) ln ( , ) z x y f x y i x y r x y    In this way the incident light and reflected light can be separated. In that the low-frequency part of the image is the incident light and the high-frequency part is reflected light, the left will be the low-frequency part (incident light) when we conduct Lowpass filtering to ( , ) z x y , as shown below: A