Image Scale Measurement with Correlation Filters in A Volume Holographic Optical Correlator
Tianxiang Zheng,Liangcai Cao,Qingsheng He,Guofan Jin
DOI: https://doi.org/10.1117/12.2035188
2013-01-01
Abstract:A search engine containing various target images or different part of a large scene area is of great use for many applications, including object detection, biometric recognition, and image registration. The input image captured in real-time is compared with all the template images in the search engine. A volume holographic correlator is one type of these search engines. It performs thousands of comparisons among the images at a super high speed, with the correlation task accomplishing mainly in optics. However, the inputted target image always contains scale variation to the filtering template images. At the time, the correlation values cannot properly reflect the similarity of the images. It is essential to estimate and eliminate the scale variation of the inputted target image. There are three domains for performing the scale measurement, as spatial, spectral and time domains. Most methods dealing with the scale factor are based on the spatial or the spectral domains. In this paper, a method with the time domain is proposed to measure the scale factor of the input image. It is called a time-sequential scaled method. The method utilizes the relationship between the scale variation and the correlation value of two images. It sends a few artificially scaled input images to compare with the template images. The correlation value increases and decreases with the increasing of the scale factor at the intervals of 0.8 similar to 1 and 1 similar to 1.2, respectively. The original scale of the input image can be measured by estimating the largest correlation value through correlating the artificially scaled input image with the template images. The measurement range for the scale can be 0.8 similar to 4.8. Scale factor beyond 1.2 is measured by scaling the input image at the factor of 1/2, 1/3 and 1/4, correlating the artificially scaled input image with the template images, and estimating the new corresponding scale factor inside 0.8 similar to 1.2.