The Hough transform (HT) is widely used in many pattern recognition problems to detect different features in the image by transforming each image pixel into parameter space using parametric equation of the shape to be detected. Digital techniques to implement HT require a very large accumulator memory and a large number of calculations and as such are limited to nonreal-time problems. To implement a real time transformation the massively parallel connectivity of neural network along with the high-speed, interference-free processing of optical systems can be applied. Previous optical HTs were limited to the switching speed of the 2D SLM. In our new implementation(Figure 1) the computation time is the aperture time of the AO cell, and is independent of the image size. Additionally there are no mechanical components, and the computation is substantially simpler than that in the other techniques. The implementation of the reconfigurable HT based on a neural network model, uses an acousto-optic cell (mercurous chloride) in shear mode. Software generates the values of the weight matrix based on the size of the image and shape to be detected. A digital-to-analog converter generates a waveform corresponds to the image vectors(see Figure 2), which is the input to the AO cell. In experiment one the weight matrix and the image vector are simultaneously presented to the LCD (Figure 3). In experiment two a simpler weight matrix was encoded on the LCD and an AO cell contained the image vector (Figure 4). Both experiments correctly detected the presence of lines in the input image.