Progressive image transmmision using dynamic binary thresholding

This paper develops a new progressive image transmission scheme. The proposed image transmission scheme is especially applicable to many interactive image communications such as telebrowsing image database, videoconferencing, or remote surveillance. While it is known to be a lengthy time to transmit high resolution images over low speed channels, a remote user may want to have an approximate image built up quickly at an early stage, he can then refine the image resolution progressively via several passes to satisfy a viewer's request. Tzou[l] presented a survey of progressive image transmission techniques including spatial domain, transform domain, and pyramid-structured approaches. According to his analysis, both transform domain and pyraid-structured domain approaches have adv.intage of low bit rate, however, the computational load is too heavy. On the contrary, several approaches in the spatial domain have the least computational load with a higher bit rate. In this paper, we want to propose another progressive image transmission technique in the spatial domain. The proposed scheme has to meet two criteria. First, it is simple in computation but it can have a bit rate improvement. Second, the selective fill-in function can be used on the transmitted image, i.e., the viewer can ask to transmit only the desired portion of an image specified by a window, thus, it avoids the waste of transmitting the unuseful data. Bas"d on these two conditions, we develop a new scheme that the location of pixel is transmitted in stead of transmitting the conventional pixel intensity. The new progressive image transmission scheme encodes an approximate binsry image initially. We apply a predefined threshold of (63,191) to reconstruct the image and transmit it to the other side. Both sides of the communication then can consistently maintain two lists of data. Each list records the locations of pixels of homogeneous intensity. However, these two lists need a lot of cost to transmit. We suggest that the communication cost of the first pass can be sharply reduced if an jppropriate compression method is applied. The reconstructed binary image is then input to the quadtree decomposition algorithm with which the output is compressed using a compression technique based on the context-free grammar[2]. So, in the first pass, the receiver can reconstruct the initial binary approximate image using the transmitted quaternary codes. Afterward, two lists of location belonging to intensities 63 and 191 are generated respectively. Based on these two lists, we apply the dynamic thresholding technique to progressively refine the image. The dynamic thresholding is defined as the mean of the pixel intensities for each list on the original image. Upon the viewer's request, the sender will generate a new list from the old one. The partitioning depends on two principals: (1). If the mean is less than and equal to the old threshold, all locations for pixels with intensities less than and equal to the mean value will be removed from the old list to create a new list. (2). If the mean is greater than and equal to the old threshold, all locations for pixf-ls with intensities greater than and equal to the mean value will be removed from the old list to create a new list.