| Summary: | 碩士 === 大同大學 === 資訊工程學系(所) === 103 === Haar-like feature extraction and Adaboost algorithm were applied to the human face detection in the beginning. Then they were widely used in the detection to other objects. In previous study, the Walsh-Hadamard transform is applied to replace the integral image and Haar-like feqtures to obtain the features of images. Through the Adaboost learning the process is used to detect human face.
In this these, the other object detection of Walsh-Hadamard transform and Adaboost algorithm is studied, including the windows of building, the license plate of cars and wheels of motorcycles. Furthermore, in addition to the original Haar-like features, the Sobel operator is used to find the gradient. The direction of the gradient is calculated to obtain the cumulative histogram. And the Haar-like features are extracted from the histogram. These three experiments are studied to evaluate the effects of the learning samples with detection results and the influence in the angle of the object.
From the experiment can be found that: less learning samples are needed with application of the Walsh-Hadamard transformation to get a good detection results. The changes in the angle of the object is less than original Haar-like features methods. Therefore, the conclusion that the object detection with Walsh-Hadamard transformation can replace the Haar-like features in the applications of objects detection.
|
|---|
