Nowadays digital multimedia data plays an important role in everyday's life. These data can beeasily generated by a variety of devices such as cell phones, cameras and so on. On the other hand, a lot of images editing softwares exist. Among these volumes of everyday publishing and broad casting images, property rights are also an important issue. Also original image detection becomes more and more difficult for humans. To help solve this upcoming issue many ways proposed. Image cropping is one of the most common ways of image changing. In this thesis, in order to recognize the original image correspond to a cropped image a smart network based on the wavelet network but with structural change and also neurons type change is presented. Our proposed smart network is a fixed grid wavelet network with the explanation that in standard type wavelet network neurons are from both wavelet function forms and scale function forms and either their combination. So our proposed fixed grid wavelet network expanding its neurons library and can approximate output with more accuracy. For proposed network neurons, an analytic wavelet family, beta wavelet family, is used. In this wavelet family, shape of wavelet and scale functions can be changed by two shape controlling parameters. In the first step, by using the orthogonal least squares method we find our dominate network neurons and determined scale and shift parameters of each selected neurons. Scale and shift parameters don’t change anymore in the following steps. Then by using least square method the network output weights are updated. After our proposed network structural formation is accomplished, we use our fixed grid beta wavelet network to identify images of a library. In our proposed method we transmit color image from RGB to HSV color form. Thus our network output is three weight vectors. In order to,correctly identify the original image corresponding to a cut image, it is necessary to classify weight vectors according to neuron function types. Each classified vector is screened with a type appropriate threshold. Simulation results show that our proposed network for recognizing up to 80 percent cut image drives to better results than other well known network in this field. It should also be noted that this network can be used for other applications, such as function approximation. Keywords: Fixed Grid Wavelet Network, Beta Wavelet Family, Feature Extraction, Cutted images
