--- title: Text2Image Parameters - READ-COOP description: "Documentation for Developers: Text2Image Parameters" --- # Text2Image Parameters ![READ-COOP](https://7528315.fs1.hubspotusercontent-na1.net/hub/7528315/hubfs/raw_assets/public/mV0_d-cms-elevate-theme_hubspot/elevate/images/avatar-placeholder.jpg?width=48&height=48&name=avatar-placeholder.jpg) READ-COOP March 1, 2021 ### thresh ``` Threshold for text alignment. If the confidence of a text-to-image alignment above this threshold, an alignment is done (default = 0.0). A good value is between 0.01 and 0.05. Note that the confidence is stored in the pageXML anyway, so deleting text alignments with low confidence can also be made later. ``` ### hyphen can be null, a non-negative double value or a json-string (default: null). If no value is set or value is “Infinity”, no hyphenation is done. If value is a positive double value, the value are the additional costs to recognize a hyphenation. The default hyphenation signs at the end of the line are ‘¬’, ‘-‘, ‘:’, ‘=’. The default hyphenation signs at the beginning of the line are empty. There can be hyphenations between all letter-pairs. If one wants to use hyphenation rules for a specific language, this can be configured using the key ‘hyphen_lang’. The hyphenation sign in the groundtruth will be ‘¬’. If one wants to configure more, one has to write a j-son-string. Keys: prefixes: list of hyphenation sign that can be hyphens at the beginning of a line (default: empty) suffixes: list of hyphenation sign that can be hyphens at the end of a line (default: empty) skipSuffix: boolean if suffix is optional (true) of forced (false) (default: false) skipPrefix: boolean if prefix is optional (true) of forced (false) (default: false) hypCosts: non-negative value that produces additional costs to recognize a hyphenation. (default: 0.0) pattern: language pattern (e.g. EN_GB, EN_US, DE, ES, FR,…) (default: empty) example: “{ “skipSuffix”:false, “skipPrefix”:true, “suffixes”:[“¬”,”-“,”:”,”\u003d”], “prefixes”:[“:”,”\u003d”], “hypCosts”:6.0, “pattern”:”EN_GB” }” one of the 4 suffixes have to be recognized and one of the both prefixes can be recognized. Hyphenation costs of 6.0 are added. Hyphenation is only possible as defined for language EN_GB. ### hyphen_lang ``` if hyphen is given, hyphenation-rules from different languages can be applied. If value = null or empty, a linebreak between all letters is possible (unicode-characters of Category L). Otherwise, a rule is applied ( see https://github.com/mfietz/JHyphenator.git for details). The language have e.g. "DE" for German and "EN" for English. Default = null. ``` ### skip_word ``` makes it possible to skip a word, for example if a baseline is too short (default: null). The value have to be a positive double value. It repesents the default delete-costs for each character. A good value is 4.0. The higher the value, the less words were skipped. If value = 0, a word can be deleted without producing costs (destroys the algorithm), if value = Infinity, no characters can be deleted. ``` ### skip_bl ``` makes it possible to skip a baseline (default: null). Sometimes the LA finds a baseline in noise (aka false positive). It is possible to delete those baselines instead of "pressing" a sequence into the line. The value has to be positive double value. The lower the value, the easier a line is ignored. A good value is 0.2. ``` ### jump_bl ``` makes it possible to handle wrong reading order in the LA (default: null) The value makes it possible to jump instead of the after a line to every other line. If value = 0, the reading order has no effect at all. If value = Infinity is the same like value = null. If you cannot trust the reading order, set value = 0. ``` ### best_pathes ``` if the number of confmats and references gets too large, one can only keep a specific number of paths at each reference. As default all paths are calculated (like setting value = Infinity). A good value is 200.0 ``` ## Share this post [mailto:https%3A%2F%2Fblog.transkribus.org%2Fen%2Ftranskribus%2Fdocu%2Ftext2image%2Ftext2image-parameters](mailto:https%3A%2F%2Fblog.transkribus.org%2Fen%2Ftranskribus%2Fdocu%2Ftext2image%2Ftext2image-parameters) ## Keep reading [![](https://blog.transkribus.org/hs-fs/hubfs/Featured%20image_multilingual%20models%20in%20Transkribus.png?width=1280&height=720&name=Featured%20image_multilingual%20models%20in%20Transkribus.png) Research Text recognition models ### Breaking the language barrier: 3 multilingual Transkribus research projects ](https://blog.transkribus.org/en/multilingual-ai-text-recognition-model-transkribus?hsLang=en) [![](https://blog.transkribus.org/hs-fs/hubfs/Text-Titan-II.jpg?width=2406&height=1632&name=Text-Titan-II.jpg) ### Introducing Text Titan II ](https://blog.transkribus.org/en/introducing-text-titan-ii?hsLang=en) ```json { "@context" : "https://schema.org", "@type" : "BlogPosting", "author" : { "@type" : "Person", "name" : "READ-COOP", "url" : "https://blog.transkribus.org/en/author/read-coop" }, "dateModified" : "2024-12-13T04:18:40.250Z", "datePublished" : "2021-03-01T23:00:00.000Z", "headline" : "Text2Image Parameters - READ-COOP", "mainEntityOfPage" : { "@id" : "https://blog.transkribus.org/en/transkribus/docu/text2image/text2image-parameters", "@type" : "WebPage" }, "publisher" : { "@type" : "Organization", "logo" : { "@type" : "ImageObject", "url" : "https://blog.transkribus.org/hubfs/transkribus_logo_long-1.png" }, "name" : "READ-COOP SCE" } } ```