Identification and tracking of dendritic spine morphology from two-dimensional time-lapsed images plays an important role in neurobiological research. Such analysis can enable us to derive a correlation between morphological characteristics and molecular mechanism of dendritic spine development and remodelling. Moreover, Neuronal morphology of hippocampal Cornu Ammonis 1 region is critical for understanding the Alzheimer's disease. Therefore, we need to extract and trace the dendritic spines accurately for examining their development and remodelling processes, which are related to functions of hippocampal Cornu Ammonis 1. There are some problems to be solved in related researches. Noise due to the properties of optical microscopes makes it difficult to identify and trace dendritic spines accurately. To solve these problems, in this paper we present a local spine detection technique minimizing noise influence in two-dimensional optical microscopy images. Also, we suggest an efficient mapping method for tracking the dynamics of dendritic spines to measure their morphological changes quantitatively. First, to utilize structural feature of spines, which are small protrusions of tree-like dendrites, we extract the tips of each dendritic branch and use this position as an initial contour position for a deformable model-based segmentation. We then use a geodesic active contour model to detect the spines accurately. Secondly, we apply an optical flow method, which takes into account both structure and movement of objects, to map every time-series image frame. Proposed method provides accurate measurements of dendritic spine length, volume, shape classification for time-lapse images of dendrites of hippocampal neurons. We compared the proposed spine detection algorithm with manual method performed by biologists and noncommercial software NeuronIQ. In particular, this method is able to segment dendritic spines better than existing methods with high sensitivity in adjacent spines and noisy images. Also the algorithm performs well compared to a human analyser.