A new character-based indexing method using frequency data for Japanese documents

Document image archives are increasingly used to replace paper and microfilm filing. Usually those archives are combined with a database management system or with full text retrieval to search the documents. An additional retrieval method to search already known images in personal document image archives using layout knowledge is presented. This knowledge can be the size, position, and color of layout objects, but also the position of keywords. A layout editor is proposed, which allows the interactive generation of a layout query using object oriented drawing functions. The layout search can be done in a relational database, which was filled with the help of layout recognition methods. For better results and performance special search methods and structures are necessary. Examples for those methods are quad trees to locate layout objects at absolute positions, neighborhood tables to find object pairs with certain spatial relationships, and full text search in page or object areas. >

Retrieval of document images using layout knowledge

Neural networks generally require significant memory capacity/bandwidth to store/access a large number of synaptic weights. This paper presents an application of JPEG image encoding to compress the weights by exploiting the spatial locality and smoothness of the weight matrix. To minimize the loss of accuracy due to JPEG encoding, we propose to adaptively control the quantization factor of the JPEG algorithm depending on the error-sensitivity (gradient) of each weight. With the adaptive compression technique, the weight blocks with higher sensitivity are compressed less for higher accuracy. The adaptive compression reduces memory requirement, which in turn results in higher performance and lower energy of neural network hardware. The simulation for inference hardware for multilayer perceptron with the MNIST dataset shows up to 42X compression with less than 1% loss of recognition accuracy, resulting in 3X higher effective memory bandwidth and ∼19X lower system energy.

Adaptive weight compression for memory-efficient neural networks

When inspecting information visualizations under time critical settings, such as emergency response or monitoring the heart rate in a surgery room, the user only has a small amount of time to view the visualization “at a glance”. In these settings, it is important to provide a quantitative measure of the visualization to understand whether or not the visualization is too “complex” to accurately judge at a glance. This paper proposes Pixel Approximate Entropy (PAE), which adapts the approximate entropy statistical measure commonly used to quantify regularity and unpredictability in time-series data, as a measure of visual complexity for line charts. We show that PAE is correlated with user-perceived chart complexity, and that increased chart PAE correlates with reduced judgement accuracy. ‘We also find that the correlation between PAE values and participants’ judgment increases when the user has less time to examine the line charts.

At a Glance: Pixel Approximate Entropy as a Measure of Line Chart Complexity

Lossless compression of magnetic resonance images is reviewed using both the theoretical and implementation models. The compression level of selected algorithms (Lempel-Ziv and Huffman) are compared against the first-order, second-order, and conditional entropies. It is found that the compression upper limit for Huffman is the first-order entropy and for Lempel-Ziv, the second-order or first-order conditional entropies. The experiments showed that the second-order and conditional entropies were lower per pixel than the first-order, suggesting a certain amount of dependencies between the adjacent pixels. As a result, the Lempel-Ziv achieved more compression than the Huffman. The first transformation (difference coding) improves the compression level by 6% for Huffman and 1% for Lempel-Ziv. In a second transformation, where images are split by their upper and lower bytes of each pixel, Lempel-Ziv performs better on the higher byte and Huffman performs better on the lower byte. >

Lossless compression of medical images

Entropy and Image Compression

Among the techniques used for retrieval of information from free-text or document databases, signature methods have proven to be more efficient in terms of storage overhead and processing speed. Signature methods, however, present the problem of “false drops” in which a document is identified but does not satisfy the user query. In the signature approaches such as Word Signature, and Superimposed Coding, the number of false drops is directly related to the hashing function selected, signature size, and number of signature buffers used for each document. Hashing functions also generate collisions, which will result in false drops. In addition, these signature methods do not take into account the length of the words or the positional information of the characters that constitute the word. The use of “Don't Care Characters” in the queries, therefore, is not possible. This paper presents a new signature approach in which the sizes of the signature files are dependent on the number of unique symbols in the alphabet, and therefore for all documents containing English text, the size is constant. The signature generated in this technique maintains the positional information of characters and therefore allows for Don't Care Characters to be used in the queries. Implementation results and comparison of this technique to the Superimposed Coding method is presented.

A new signature approach for retrieval of documents from free-text databases

This paper deals with both the theoretical and the practical aspects of information preserving image data compression The concept of entropy and the common effective compression algorithms are first reviewed The first-order, second-order and the conditional entropies of images are calculated and compared to the compression level of some existing algorithms A set of reversible transformation techniques are then introduced and applied to the referenced images The purpose of these transformations is to lower the entropy in order to improve the compression ratio of the compression algorithms The effect of the transformations on the compression algorithms are examined Finally, suggestions for further research in lossless compression are given© (1991) COPYRIGHT SPIE--The International Society for Optical Engineering Downloading of the abstract is permitted for personal use only

Information-preserving image data compression

Three primary processes are utilized in most pattern recognition systems. (1) The representation process in which the raw digitized data is mapped into a higher level form. (2) The storage process which generates a known base containing the high-level representation of all known patterns in a problem domain. (3) The identification process which classifies the unknown pattern, given its high-level representation and the known base. This paper introduces a new and unified approach to character recognition in which the generation of the known base and the identification process are independent of the representation process. The concept of learning is applied where additional known data can be added to the known base thus allowing for better recognition. The concept of fuzziness is applied to the identification process to find the best match. Algorithms have been developed and applied to digitized characters. The processes of representation, storage, and identification used for this implementation along with results are discussed.© (1992) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

Character recognition: a unified approach

The degree to which digital images are recognized correctly by computerized algorithms is highly dependent upon the representation and the classification processes. Fuzzy techniques play an important role in both processes. In this paper, the role of fuzzy representation and classification on the recognition of digital characters is investigated. An experimental Neural Network model with application to character recognition was developed. Through a set of experiments, the effect of fuzzy representation on the recognition accuracy of this model is presented.

/pdf/character-recognition-using-a-neural-network-model-with-1iefc2d5ra.pdf

Nassrin Tavakoli

Papers

Entropy and Image Compression

A new signature approach for retrieval of documents from free-text databases

Information-preserving image data compression

Character recognition: a unified approach

Character recognition using a neural network model with fuzzy representation