Stereo vision is fast becoming a highly investigated area in the domain of image processing. Depth information may be obtained from stereo or multi-vision images for reconstructing objects in 3D based on 2D information. Robotic applications make use of stereo vision for navigation purposes, locking down targets, as well as simulating human-like behaviour. This paper presents an algorithm for the auto-alignment of stereo images followed by the self-extraction of objects of interest using an unsupervised search. Based on the understanding that different objects or regions are focused at different focal points, alignment between the two images is carried out to determine areas of high overlapping similarities. Objects are then identified in these selected areas with their estimated depth calculated based on the disparities between the stereo images. Results obtained for tests carried out on several experimental image pairs showed good extraction of the objects with close-to-real-world values obtained for the distances of the objects from the cameras.