The dynamics of semantic information in biosystem is studied based on holons, generators of mutual relations. Any biosystem has an internal world, a so-called "self", which has an intrinsic purpose rendering the system continuously alive and developed as much as possible against a fluctuating external world. External signals to the system through sensory organs are classified by the self into two basic categories, semantic information with some meaning and value for the purpose and inputs from background and noise sources. Due to this breaking of semantic symmetry, any input signals are transformed into a figure and background, respectively. As a typical example, the visual perception of vertebrates is studied. For such semantic transformation the external signal is first decomposed and converted into a number of elementary signs named "syntons" which are then transmitted into a sensory area of cortex corresponding to an image synthesizer. The synthesizer is a sort of autonomic parallel processor composed of autonomic units, "holons", which are characterized by many internal modes. Syntons are fed into the holons one by one. A set of the elementary meanings, the so-called "semons", provided to the synton are encoded in the internal modes of the holon; that is, each internal mode encodes a semon. A dynamic information theory for the transformation of external signals to semantic information is developed based on our model which we call holovision. Holovision is a dynamic model of visual perception that processes an autonomic ability to self-organize visual images. Autonomous oscillators are utilized as the line processors to encode line elements with specific orientations in their phases as semons. An information space is defined according to the assembly of holons; the spatial plane on which holons are arranged is a syntactic subspace while the internal modes of the holons span a semantic subspace in the orthogonal direction. In this information space, the image of a figure is self-organized - as a sort of spatiotemporal pattern - by autonomic coordinations of the holons that select relevant internal modes, accompanied with compression of irrelevant syntons that correspond to the background. Holons coded by a synton are relevantly connected by means of coherent relations, i.e., dynamic connections with time-coherence, in order to represent the image that varies in time depending on the instantaneous state of the external object. These connections depend on the internal modes that are cooperatively selectively selected by the holons. The image is regarded as a bridge between the external and internal world that has both external and internal consistency. The meaning of the image, i.e., transformed semantic information, is spontaneously transferred from semantic items that have a coherent relation with the image, and the external signal is perceived by the self through the image. We demonstrate that images are indeed self-organized in holovision in the previously described sense. Simulated processes of the creation of semantic information in holovision are shown to display typical features of the forgoing steps of information compression. Based on these results, we propose quantitative indices that represent the value of semantic information in the image processor as well as in the memory.