Target Extraction Method For Pigs Based On Adaptive Ellipse Block And Wavelet Edge Detection

In order to extract pigs target form complex background overlooking pigs image,this paper proposed a target extraction algorithm based on adaptive elliptical block and wavelet edge detection, laid the foundation for pigs’ identification and behavior analysis.For the adverse situation in large-scale pig farms, such as: much noise interference, poor environment, the complex light and low quality of pigs’ image,firstly this paper used restrict contrast histogram equalization to enhance the image,improving similarity in the same area and contrast in the different area, at the same time, retaining the image details; secondly this paper used two dimensional OTSU threshold to initially segment the image, traversing the grey level to find the best threshold that made the maximum variance between target and background, using eighty pixels plate element morphological closing operation to eliminate plate strip edges and connected region that had smaller pixels, populating the prospect empty goal that didn’t belong to the prospect target. After the initial pigs target image segmentation, the original image was blocked adaptively, firstly, the least square method based on algebraic distance was used to fit the image for ellipse fitting;secondly, then the least square method based on geometric distance was used to obtain the improved parameters of ellipse, lastly, the original image was adaptive divided into several oval area for the center with each pig goal. Finally, in each oval blocks,the wavelet transform multi-scale analysis ability combined with the large scale’s good noise resistance, edge stability and small scale’s high edge location accuracy,rich details, the wavelet edge detection algorithm based on B-spline was used to find pigs target edge. After edge detection, edge growth was used to deal with not closed edge, binary morphology processing was used to remove the connected regions that was not belong pigs target and fill the cavity that was not belong prospect in pigs target, then the pigs target were extracted form the image.Test results showed that this paper method can quickly and efficiently extract targets from overlooking pigs image. Compared with the OTSU thresholdssegmentation, this paper’s three target extraction evaluation parameters included:regional gray contrast(G), regional uniformity(U) and split the cost factor(Q), were superior to the former method. Compared with the artificial target extraction, the error segmentation rate in this paper was 2.9%, the target extraction result was satisfactory and the expected effect was achieved.