| Summary: | 碩士 === 國立中央大學 === 資訊工程學系 === 102 === Sound recognition has become an important application in some devices. The type of sound to be recognized may vary, e.g., musical instrument sounds, environmental sounds, and speech. In this study we use environmental sound for our experiment.
Time-frequency, which can represent an audio signal, is a form of texture image that can be used for image classification. In this paper, we introduce a simple image classification method using local binary pattern (LBP) and an image smoothing method prior to feature extraction to reduce spectrogram image noise.
In this thesis, we combine spectrograms and LBP uniform with an image filter and variance measure (VAR) for contrast enhancement. We alsointroduce adynamic LBP method to reduce the dimension in difference dimension for each sub-band(high, middle, and low frequency). After using image filter as pre-treatment and VAR for contrast enhancement, weconcatenate all thesefeatures.
To remove image noise, we use two types of smoothing filter:a box filter (mean filter) and a Gauss filter. To improve recognition, filtering is applied as a pretreatment prior to feature extraction. To enhance local image texture contrast, such as object edges and corners, we use a VAR function. We use a support vector machine for the classifier.
|
|---|
